It’s called InventNext AI.

It’s a tool that lets anyone describe a their invention idea, answer a few follow-up question, and generate a high quality draft provisional patent application — including figures.

All in about 5–10 minutes.

If you’re not an intellectual property lawyer, that might not sound particularly dramatic.

If you are an IP lawyer, you understand exactly why it is.

Because the hardest part of the patent process typically isn’t the law.

It’s getting the invention written down.

⸻

Most Inventions Die Before Anyone Even Tries to Patent Them

People imagine the barrier to patents is technical complexity.

It’s usually not.

The barrier is inertia.

Someone has an idea.

They think it might be interesting.

Maybe even commercially valuable.

But then they hit the first question:

“Where do I even start?”

A proper patent disclosure requires:

• describing the system

• identifying the technical components

• explaining how it works

• outlining alternative embodiments

• structuring a specification

That’s intimidating if you’ve never done it before.

So most people don’t even start.

And a lot of potentially valuable inventions simply disappear.

⸻

The Patent System Was Designed for Professionals

For decades, the patent system has operated on a simple assumption: if you want a patent, you will start by hiring a professional (a patent attorney).

Someone who was trained to translate your invention idea into the structured technical disclosure required by the patent system.

That assumption shaped the entire process. And of course, professional guidance remains extremely valuable.

But one challenge has always existed before that point: turning an initial idea into a clear written description of the invention.

Generative AI makes it possible to help people with that early step.

Large language models are particularly good at organizing information and turning rough ideas into structured explanations.

That means new tools can help inventors document their ideas and explore the patent process earlier, even before deciding whether to formally pursue protection with professional counsel.

What the System Does

InventNext AI works through a guided invention interview.

First you describe your idea in plain English.

The system then asks follow-up questions — the kinds of questions a patent attorney would ask when trying to understand an invention.

Things like:

• What problem does this solve?

• What components are involved?

• How does the system operate?

• What variations could exist?

Then the system uses those responses (and any documents you attach) to generate a structured provisional patent draft.

Including:

• title and field of invention

• background and summary

• detailed technical description

• alternative embodiments

• diagrams illustrating the system architecture

The goal isn’t to replace lawyers.

The goal is to make the first step accessible.

Because once an invention is written down, the entire conversation changes.

⸻

The First Test Case Was… A Dog Translator

One of the early testers was a friend of mine.

He and his teenage daughter decided to try the system one evening.

They entered an idea that was half serious and half joking:

A device that could translate communication between humans and dogs.

They described the idea.

Answered a few questions.

And about fifteen minutes later the system produced a 19-page draft provisional patent application.

With system diagrams.

With architecture descriptions.

With claim concepts.

Now — is a dog translator a great invention?

Maybe.

Maybe not.

That’s not the point.

The point is that two people with no patent experience were able to turn a vague idea into a structured invention disclosure in minutes.

A few years ago that simply wasn’t possible.

⸻

AI Is Starting to Break Professional Bottlenecks

If you follow my writing, you know I’ve been saying for a while that AI is going to change how professional services work.

Law included.

Not because lawyers are unnecessary.

But because many parts of professional workflows are essentially structured reasoning plus documentation.

AI systems are very good at that.

They’re not a substitute for human judgment.

Or strategy.

Or responsibility.

Which is why professional review still matters.

But the idea that every single step of a process must be performed by a professional is starting to break down.

And invention disclosure is a perfect example.

⸻

This Is Just the Beginning

InventNext AI doesn’t replace the patent system.

And it doesn’t replace patent lawyers.

What it does is make the starting point much easier. And available to anyone.

Instead of staring at a blank document wondering how to describe an invention, someone can answer a few targeted questions and generate a draft.

From there they can:

• refine the invention

• evaluate patentability

• consult an attorney

• file a provisional

Or decide the idea wasn’t worth pursuing after all.

But at least the invention exists on paper.

And that’s where innovation actually begins.

⸻

The App Is Live

If you’ve ever had an idea you thought might be an invention — even a strange one — you can try it.

Just describe the idea and see what happens.

Sometimes the most interesting things start that way.

You can try it here:

inventnext.ai

And if you do test it, I’d genuinely love to hear what you think.

— Ana

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Guess which one’s in trouble.

This week exposed how completely broken the accountability framework around AI actually is. If you’re building on these platforms, using them, or just trusting that someone is making sure they’re safe — what happened should change that assumption.

OpenAI Had the Red Flags. It Sat On Them.

On February 10, an 18-year-old killed eight people in Tumbler Ridge, British Columbia — her mother, her half-brother, five students, and an educational assistant — before killing herself. One of the worst mass shootings in Canadian history.

OpenAI’s automated systems flagged the shooter’s ChatGPT account in June 2025 for violent content. The company banned it. According to the Wall Street Journal, about a dozen employees debated whether to contact police. Some pushed to report. Leadership said no.

No law required OpenAI to call anyone. So it didn’t.

The shooter made a second account and kept going. OpenAI didn’t catch it until police publicly identified her. The company has since told the Canadian government that under its new protocols, the account would have been referred to law enforcement. That’s not an apology. That’s an admission the old system failed.

Anthropic Drew Two Lines. The Pentagon Wants Them Gone.

Anthropic makes Claude — the only frontier AI model on the Pentagon’s classified networks. They asked for two restrictions: no mass surveillance of Americans, no fully autonomous weapons.

The Pentagon wants both removed. Earlier this week, Defense Secretary Pete Hegseth told CEO Dario Amodei that if Anthropic doesn’t agree to unrestricted “all lawful purposes” usage by 5 pm on Friday February 27, he would invoke the Defense Production Act — a Korean War-era coercion statute never used against a software company — and label Anthropic a “supply chain risk.”

Anthropic rejected the offer. Amodei, on the record: “These threats do not change our position: we cannot in good conscience accede to their request.” He added: “Those latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security.”

Pentagon Undersecretary Emil Michael responded by calling Amodei a “liar” with a “God-complex.” On X. In public.

Elon Musk’s xAI already signed a deal with no restrictions. Google is reportedly close. Anthropic is the last holdout.

But something is happening on the other side. Over 300 Google and OpenAI employees signed an open letter titled “We Will Not Be Divided,” urging their companies to hold the same red lines. Google DeepMind’s Chief Scientist Jeff Dean publicly backed Anthropic, posting that mass surveillance “violates the Fourth Amendment” and is “prone to misuse for political or discriminatory purposes.” Retired Air Force General Jack Shanahan said “painting a bullseye on Anthropic garners spicy headlines, but everyone loses in the end.”

The Pentagon’s strategy of isolating Anthropic is backfiring inside the very companies it’s trying to recruit.

Same Week. Opposite Consequences.

OpenAI had behavioral red flags from a user who later committed mass murder and chose not to act. It gets to write a letter to Canada promising to do better.

Anthropic is maintaining two safety guardrails and refusing to remove them. It’s being threatened with blacklisting, forced compliance under a wartime statute, and a public smear campaign from a senior Pentagon official.

AI safety in 2026 is not enforced by law. It’s enforced by whichever CEO is willing to lose money. And the system is designed to make sure they stop.

What the Law Actually Says (And Doesn’t)

There is no U.S. or Canadian federal law requiring AI companies to report dangerous user behavior to law enforcement. OpenAI’s “credible and imminent” threshold is an internal policy it wrote for itself. No statute compelled them to act. No statute penalizes them for doing nothing.

Compare that to industries where this was solved decades ago. Therapists have a legal duty to warn under Tarasoff v. Regents, 1976. Banks must file Suspicious Activity Reports — over 370,000 in 2024 alone. AI companies? Zero mandatory reporting obligations.

The counterargument is real: mandatory reporting could mean users get police at their door over a creative writing prompt. But the current system isn’t cautious balance — it’s nothing. There’s a wide margin between “report every edgy prompt” and “watch someone rehearse gun violence for days and shrug.”

The “supply chain risk” designation on the Anthropic side is reserved for foreign adversaries like Huawei and Kaspersky. Anthropic’s safety restrictions exist only in contract language. The only thing standing between Claude and unrestricted military deployment is one company’s willingness to take the hit.

Where the Lawsuits Land

OpenAI didn’t just ignore the shooter’s account. It flagged it, reviewed it, evaluated the risk, and made a judgment call not to report. That sequence creates real legal exposure. Under tort law, when you voluntarily undertake a duty — even one the law doesn’t require — and perform it negligently, you can be liable for the harm that follows. That’s the undertaker doctrine.

OpenAI built a content moderation system. It flagged violent content for human review. Once its employees were staring at gun violence scenarios and debating whether to call the police, the company was making an affirmative safety judgment. If the families of those victims file a wrongful death suit — and I’d be surprised if they don’t — that internal deliberation becomes the centerpiece of discovery. Every email, every Slack message, every policy memo. Proving causation won’t be simple — plaintiffs would need to show a report would have changed the outcome, and that’s a high bar. But OpenAI’s own diligence may become the evidence that it fell below the standard of care it set for itself. That’s the kind of fact pattern that survives a motion to dismiss.

On the Anthropic side, the supply chain risk designation wouldn’t just kill a Pentagon contract. Claude is embedded across the enterprise market — through Anthropic directly, through Amazon Bedrock, through Palantir. If the Pentagon pulls the trigger, every company that uses Claude and does business with the federal government would need to certify Anthropic’s technology is nowhere in their stack. Not startups. Fortune 500 companies, defense contractors, financial institutions, every federal vendor in between. If you’re a GC at any of those companies, the question is simple: does your vendor agreement account for a government-imposed supply chain designation, and what’s your exit plan if Claude becomes toxic to your federal business?

The Bipartisan Alarm

Republican Senator Thom Tillis — not running for reelection, nothing to gain — said the Pentagon is handling this “unprofessionally” and that Anthropic is “trying to do their best to help us from ourselves.” Democrat Elissa Slotkin said at a hearing: “The average person does not think we should allow weapons systems to get into war and kill people without a human being overseeing that.”

When senators from opposite parties agree without coordinating, that’s a signal.

What Needs to Happen

Congress needs a federal AI reporting standard — defined thresholds modeled on SARs in banking. Not “report everything.” A real standard with real consequences.

Procurement policy needs a safe harbor for companies maintaining safety guardrails. “Keep restrictions and lose your contract” isn’t a choice. It’s coercion.

And safety architecture for military AI cannot live in contract language one administration can rip up. Codify it or accept it doesn’t exist.

In 2026, AI safety isn’t law. It’s a bet one company is making with its own money — and the whole system is built to make sure that bet doesn’t pay off.

━━━

Sources include the Wall Street Journal, Bloomberg, CBC News, the Associated Press, Axios, NBC News, CNN, CBS News, NPR, CNBC, Forbes, Lawfare, and the New York Times. Some details — particularly regarding OpenAI’s internal deliberations — rely on anonymous sourcing. Statements from OpenAI, Anthropic, and Pentagon officials are from on-the-record communications, public letters, and social media posts. Shooting facts confirmed by the RCMP.

━━━

📬 If you’re a founder, executive, or GC trying to figure out how to use AI without accidentally creating a compliance nightmare, that’s the work I do.

Let’s talk before you are the next one under scrutiny.

Book a call with me here: https://calendly.com/analaw/consult

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

First, Think About The World We Already Live In

Think about how you handle legal issues today. You get a letter from opposing counsel, or your GC sends you an update on pending litigation, or you’re dealing with an employee situation that might turn into a lawsuit. You type up your version of what happened so you can send it to your lawyer. You Google some of the legal terms to understand what you’re dealing with. You email your lawyer from your personal Gmail or your company Outlook. Maybe you hop on a Zoom call to talk it through.

At every step, your information is flowing through third-party platforms. Google is processing your search queries. Gmail is storing your privileged emails. Zoom is carrying your privileged calls. And here’s the thing most people don’t realize: AI is now embedded in all of these tools. Gemini is reading your Gmail and summarizing your emails. Copilot is analyzing your Outlook attachments. Zoom’s AI companion is transcribing your attorney calls. Slack’s AI is indexing your messages. You probably didn’t opt into most of this. Many of these features were turned on by default.

Every one of these platforms has terms of service that allow them to collect, process, and in some cases disclose your data. And no one—no court, no bar association, no one—has ever suggested that using Gmail or Zoom for privileged communications waives your legal protections. The legal profession runs on these tools. Privilege survives.

But a federal judge just ruled that if you take that same information and type it into ChatGPT or Claude—asking the AI to explain a legal concept, help you understand a contract, or break down what your lawyer’s memo means—that might be different. That might be a “disclosure to a third party” that destroys your privilege.

The Case Everyone Is Talking About

Bradley Heppner, a former CEO charged with $150 million in securities fraud, hired a major law firm (Quinn Emanuel) and then did what a lot of people would do: he opened Anthropic’s AI chatbot Claude and started researching his legal situation on his own. He explored legal theories, researched potential defenses, and generated about 31 documents. Then he sent them to his lawyers.

The FBI raided his house, found the AI documents on his devices, and the government wanted them. His lawyers said they were protected. On February 10, 2026, Judge Jed Rakoff in the Southern District of New York ruled: not protected. Hand them over. The written opinion came out February 17. It’s the first federal ruling to address this question directly.

The ruling involves two separate legal protections. I’m going to explain both, because understanding the difference matters for what you should actually do.

Protection #1: Attorney-Client Privilege

Attorney-client privilege is the rule that says your private communications with your lawyer are protected. The other side in a lawsuit can’t force you to turn them over. The government can’t read them. It’s one of the most fundamental protections in the legal system. But it only works if certain conditions are met: you need an actual attorney on the other end, the communication needs to be for the purpose of legal advice, and it needs to be kept confidential.

The court said Heppner’s AI interactions failed every one of those conditions. Claude isn’t a lawyer—it can’t form an attorney-client relationship with anyone. Its own terms say it doesn’t provide legal advice, so you can’t claim you were seeking legal advice from it. And the consumer version’s privacy policy says Anthropic can collect your inputs, train its models on them, and disclose them to government authorities. No confidentiality, no privilege.

On these specific facts, the court’s reasoning is solid. Heppner wasn’t talking to a lawyer. He was using software. And you can’t create privilege retroactively—making documents on your own and then sending them to your lawyer doesn’t make the documents privileged after the fact.

But Judge Rakoff left the door open. He said that if the platform did have confidentiality protections—like an enterprise agreement—and the person was acting on their lawyer’s instructions, the analysis could flip entirely. Essentially: use the right platform and have your lawyer direct the work, and AI interactions might be protected.

Here’s the Part That Should Worry You

Heppner didn’t just type his own thoughts into Claude. He took information his lawyers had given him and pasted it into the chatbot. The court found that by putting privileged attorney-client communications into a consumer AI tool, he may have destroyed the privilege on the original communications from his lawyers—not just the AI-generated documents.

Read that again.

Your lawyer sends you a strategy memo. You paste it into ChatGPT because you want to understand what it’s saying—maybe the legal language is confusing, maybe you want it simplified. Under this ruling, you may have just blown up the privilege on your lawyer’s actual advice to you. The original communication. Gone. Because you used a chatbot to read it.

Now, does that make sense? Think about what people actually do when they’re dealing with a legal issue. You write up what happened—a timeline, a statement, the details you think your lawyer needs to know. You Google unfamiliar terms. You might ask a trusted friend or your CFO what they think. You email your notes to your lawyer through Gmail. Your thought process is embedded in every step of that workflow, and every step runs through a third-party platform.

Using AI is the same activity. When you open Claude and type “what does ‘breach of fiduciary duty’ mean” or “help me organize what happened in chronological order” or “explain my lawyer’s memo to me in plain English,” you’re doing what people have always done: trying to understand your situation so you can work with your attorney effectively. You’re not “disclosing” privileged information to a third party. You’re using a tool to participate in your own legal matter.

Here’s a test. Your lawyer sends you a privileged memo. You upload it to Google Translate because you think better in another language. You just submitted the entire contents of a privileged communication to a third-party platform whose terms allow data collection and disclosure. Under the Heppner reasoning, that’s a waiver. But nobody believes that, because everyone understands that using a translation tool is just using a tool—not sharing secrets with Google.

Paste that same memo into ChatGPT and say “explain this in plain English”—same purpose, same intent, different tool—and suddenly it’s a waiver? The people most exposed by this reasoning are the ones who need the most help understanding their lawyers. That’s not a framework that protects the attorney-client relationship. It’s one that penalizes people for trying to participate in it.

This creates a doctrinal inconsistency that the opinion doesn’t address. Every time you send a privileged email through Gmail, you voluntarily put that communication on Google’s servers. Google stores the full text. Google’s AI now reads it, summarizes it, generates responses based on it. Google’s privacy policy allows use and disclosure. Under the Heppner reasoning, that’s a voluntary disclosure to a third party without confidentiality guarantees—the same act, the same terms, the same AI processing. But no court has ever called it waiver. The legal system treats email platforms as tools. The ruling treats chatbots as third-party recipients. That distinction has no doctrinal basis—it’s a difference in interface, not in law.

Protection #2: Work Product

The second protection is called work product. It’s separate from privilege and works differently. Work product protects the materials that get created when someone is preparing for litigation—the research, the notes, the strategy documents, the drafts. The strongest version, “opinion work product,” protects an attorney’s mental impressions—their strategic thinking about the case. The basic principle, from a Supreme Court decision called Hickman v. Taylor, is that the legal system breaks if one side can just take the other side’s preparation and use it against them.

Heppner lost on work product because, in his court’s jurisdiction, these materials generally need to be prepared by a lawyer or at a lawyer’s direction. His lawyers admitted they never told him to use Claude. He did it entirely on his own. So the court said: not protected.

The court framed work product’s purpose narrowly—as protecting “lawyers’ mental processes.” But the original Supreme Court decision said something broader: it’s about preventing the other side from borrowing “the wits of the other side.” That’s not limited to lawyers. It’s about whether your opponent gets to freeload off your preparation—whoever did the preparing.

And think about what Heppner was actually doing. He’s facing massive criminal charges. He sits down with AI and works through the law, tests defense theories, tries to figure out how the government’s case might work. He’s preparing for litigation. If he’d done the exact same thinking with a yellow legal pad—five drafts, testing ideas, crossing things out, starting over—those notes would almost certainly be protected. His strategic choices and mental impressions would be embedded in every page. Using AI to do that same thinking—prompting, revising, trying a different angle, discarding one approach and testing another—is the same cognitive process. The tool is different. The thinking is identical.

But the court didn’t look at it that way. It asked where the data went instead of what the person was doing. There’s already a federal court that took the opposite approach: in Tremblay v. OpenAI (N.D. Cal. 2024), a judge classified attorneys’ ChatGPT prompts as the highest level of protected work product, because the prompts reflected the lawyers’ strategic thinking. The key difference is that Tremblay involved lawyers and Heppner involved a client acting on his own. But the line between those two is thinner than it looks—especially when a client is doing AI research at their lawyer’s direction, which is already happening.

How courts resolve this going forward—do we ask “what was the person doing?” or “where did the data go?”—is going to shape AI and legal protection for the next decade.

What the Court Got Wrong About How AI Actually Works

What the court thinks “training on your data” means vs. what’s actually happening.

The court cited the fact that consumer Claude “trains on user data” as a key reason there’s no confidentiality. The implication is that Anthropic receives your inputs, reads them, and holds onto them—like a person you handed a document to. That’s not what’s happening. Not even close.

When an AI model is “trained” on your data, it doesn’t save your data anywhere. Training is a mathematical process that adjusts billions of numerical values inside the model. Your input nudges the model’s general patterns in some infinitesimal way, but the actual content of what you typed—your words, your questions, your documents—gets dissolved. Think of it like pouring a cup of water into the ocean. The ocean changes in some unmeasurable way. Nobody is fishing that specific cup back out.

If you asked a future version of Claude “what did Bradley Heppner type about his legal defense,” it would have absolutely no idea. It can’t retrieve his inputs. It can’t reproduce his documents. AI researchers have found that models can occasionally memorize fragments of highly unusual or frequently repeated training data, but 31 documents from one user in a dataset of billions? The chance of meaningful reproduction is essentially zero.

This matters because the whole concept of privilege waiver assumes that when you share information with a third party, that third party has it—they can remember it, repeat it, be called to testify about it. After AI training, nobody has it. No person at Anthropic read it. No system can reproduce it. The “third party” can’t be brought into court and asked what you said.

Now, waiver doctrine doesn’t technically require the third party to be able to reproduce what you shared—it turns on whether your disclosure was voluntary and inconsistent with maintaining confidentiality. But that standard evolved for a world of human recipients, where disclosure to a third party meant a person now possessed your information. AI creates a category the doctrine never contemplated: a process that ingests information without any person or system meaningfully receiving it. The court applied the old framework without acknowledging that the underlying assumptions don’t map onto this technology.

There’s an important distinction here: training is separate from data storage. Anthropic does store your actual conversation logs on its servers for a period of time, and those stored conversations could be subpoenaed. That’s a real risk. But the court treated training itself as an independent reason to destroy confidentiality—and that’s treating a mathematical process that makes your data lessaccessible as if it were the same as handing someone a copy of your documents.

The opinion completely ignores temporary chats.

Both Claude and ChatGPT have temporary or incognito modes where the conversation isn’t saved to your account and, according to the platforms, isn’t retained the way normal conversations are. The opinion doesn’t address this at all. If Heppner had used a temporary chat, there likely would have been nothing on his devices for the FBI to find and nothing on Anthropic’s servers to hand over.

So what then? Could the government force him to describe from memory what he asked the AI? In a criminal case, the Fifth Amendment almost certainly prevents that. In a civil case, he could theoretically be asked about it in a deposition—but that testimony would sound like: “I was trying to figure out whether a certain defense might work, so I asked something along these lines, and the AI said something like this.” That’s not describing a conversation with another person. That’s someone describing their own thought process about their own case. Which is exactly what work product exists to protect.

Under the Heppner framework, a court might say: it doesn’t matter that the record is gone—the waiver happened when you typed it. But that would mean the information has vanished, no one has it, the AI can’t reproduce it—and you’ve still permanently lost the right to protect your own thinking because you once typed it into a text box. That result is difficult to square with the purpose of waiver doctrine, which exists to address the risk that a third party will use or disclose what you shared—not to penalize the act of using a particular tool.

Google Search is the same thing in a different box.

When you Google “elements of securities fraud,” your query goes to Google’s servers. Google stores it. Their terms allow use and disclosure. Under the Heppner reasoning, that’s a disclosure without confidentiality. But no one has ever argued that Googling legal questions waives anything. And in 2026, Google has AI built into search—when you type a legal question, an AI model processes your query and writes you a natural language answer. Functionally, that’s the same thing ChatGPT does. The only difference is that one looks like a search bar and the other looks like a chat window. A difference in interface design shouldn’t determine whether you keep your legal protections.

The Subscription Tier Trap

If you’re thinking “I pay for the premium version and I turned off training, so I’m fine”—you’re not. Turning off the training setting means your inputs won’t be used to improve future versions of the model. That’s all it does. The privacy policy—which is what the court actually looked at—still allows the platform to hand your data to the government. Toggling a setting is not the same as having a confidentiality agreement.

Here’s the distinction that actually matters: consumer AI is a personal productivity tool. Enterprise AI is an evidentiary environment. They look identical—same interface, same models, same text box. But the legal infrastructure underneath is completely different. Consumer plans run on terms of service that don’t guarantee confidentiality. Enterprise plans run on negotiated commercial agreements with contractual confidentiality protections, restrictions on data use, and limits on disclosure. One is a convenience. The other is a legal architecture. And right now, most companies are running sensitive work through the convenience version.

Here’s something almost nobody knows: Anthropic’s Team plan, at $30 per user per month, is actually governed by their Commercial Terms—the same legal category as Enterprise. It’s not a consumer product, even though it’s sitting right next to the consumer plans on the pricing page. Whether those terms would actually hold up in court to preserve privilege is untested. But it’s a fundamentally different legal position than a $200/month consumer subscription with zero confidentiality.

Companies that get this right now gain a real competitive advantage: they can move faster with AI without creating litigation exposure. Companies that don’t are building a discoverable record of every sensitive question anyone on their team has ever asked a chatbot.

But there’s a harder version of this problem. Enterprise plans can cost $50,000 or more per year. If your legal protections depend on which AI subscription you can afford, we’ve built a system where well-resourced companies get to use AI with legal protection and everyone else is exposed. That includes your employees dealing with personal legal issues on the same AI tools they use at work. That includes the small business owner who can’t afford enterprise pricing. That includes anyone who doesn’t have a legal team telling them which text box is safe and which one isn’t. That’s a gap that should make everyone uncomfortable.

What You Should Do Right Now

Before the specifics, a quick vocabulary check—because these terms get thrown around and most people use them interchangeably, but they mean different things.

• Not privileged means the protection never existed in the first place—the communication didn’t meet the legal requirements.

• Waived means the protection existed but you destroyed it by doing something inconsistent with keeping it confidential—like pasting your lawyer’s memo into a consumer chatbot.

• Discoverable means the other side in litigation can demand it and you have to hand it over. The Heppner ruling implicates all three, and each creates a different kind of exposure.

Here’s a scenario to make this concrete. You’re in the middle of an acquisition. Your GC sends you a memo flagging potential antitrust exposure in the target’s pricing practices. You paste that memo into ChatGPT and ask it to summarize the key risks so you can brief the board. Under Heppner, you may have just waived privilege over your GC’s original analysis—and opposing counsel in any future litigation over that deal could demand both the AI conversation and the underlying memo. Your board-level strategy session just became someone else’s evidence.

If you’re an executive or founder: Audit how your company is using AI today. If anyone on your team is using consumer ChatGPT or Claude for anything that touches legal strategy, regulatory questions, HR disputes, or litigation preparation, those conversations are potentially discoverable by the other side. That means opposing counsel in a lawsuit could demand them. Get your team on enterprise agreements with real confidentiality protections, or create a clear policy about what can and can’t go into consumer AI tools.

If you’re a GC who thinks your current AI policy covers this: It probably doesn’t. Most corporate AI policies say something like “do not enter confidential information into AI tools.” That’s a start, but it doesn’t define what counts as confidential, it doesn’t distinguish consumer from enterprise platforms, it doesn’t address the waiver risk when employees paste your legal memos into a chatbot, and it doesn’t account for the AI features now embedded in the email and collaboration tools your entire organization already uses. A policy that says “don’t put sensitive stuff in ChatGPT” while Gemini is reading every email in your company’s Gmail isn’t a policy. It’s a false sense of security.

If you’re personally dealing with a legal matter: Do not paste your lawyer’s communications, case-specific details, or identifying information about your legal situation into consumer AI tools. If you want to research legal concepts, ask general questions in general terms—“what is a breach of fiduciary duty” is fine, “here’s what my company did, what’s my exposure” is not. The friendly chat interface feels private. It’s not.

If you work with a lawyer: Ask them whether they have a policy on AI and privilege. If they don’t, they should. And if they’re using consumer AI for work on your case—their own research, their own strategic analysis—ask what platform they’re on and what the terms say about confidentiality. A different federal court has already recognized that lawyers’ AI prompts are protected work product (Tremblay v. OpenAI), but that argument is much stronger on an enterprise platform than on consumer terms.

If you’re a lawyer reading this: Tell your clients, in writing, that consumer AI is not confidential. Put it in your engagement letter. Warn them that pasting your communications into a chatbot could destroy yourprivilege, not just theirs. If you’re directing clients to use AI, document that direction—it could be the difference between protected and discoverable.

━━━

🧠 Bottom line: This is a district court ruling. Not binding on other courts. But it’s from one of the most prominent federal courts in the country, from a well-respected judge, and it will be cited in litigation everywhere. Within twelve months, expect plaintiff firms to start adding AI-prompt production requests to standard discovery. This is coming.

The reasoning treats chatbots as a special category of disclosure—while ignoring that AI is already embedded in the email, search, video, and collaboration tools the entire profession uses every day without anyone suggesting privilege is waived. That doctrinal inconsistency will have to be resolved. Until it is, the instability itself is the risk. And even where privilege isn’t technically waived, these AI interactions may still be independently discoverable.

Consider this your written warning. AI is now part of your litigation surface area. Every prompt you type could end up on the other side’s exhibit list. Treat it accordingly.

━━━

Executive Summary: What This Means for You

• Consumer AI use touching legal issues = privilege waiver risk.

• AI-generated documents and logs are potentially discoverable—even if privilege isn’t waived.

• Your internal AI policy likely doesn’t address privilege waiver.

• Enterprise contracts with confidentiality protections change the legal analysis.

• Toggling “training off” does not create confidentiality.

• Opposing counsel will test this aggressively in discovery. Plan for it now.

━━━

📬 If you’re a founder, executive, or GC trying to figure out how to use AI without accidentally creating a discovery nightmare, that’s the work I do.

Let’s talk before your prompts become someone else’s evidence.

Book a Legal Risk Consult: https://calendly.com/analaw/consult

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

My inventions are an engineered system that every AI lab will need so their AGI can be verified, adaptively smart, and safe.

We all know AI is everywhere. AGI is AI that can match or beat human-level intelligence… basically the unicorn Silicon Valley has been chasing.

VSI is where it starts.

The first patent I filed, Verified Synthetic Intelligence (VSI), is the foundation that every credible AI—and every true AGI—will need.

It makes machines capable of proving—and verifying—how they think.

With VSI, every reasoning step leaves a trace—a cryptographic and causal signature that proves how a conclusion was reached. In human terms, it’s like keeping an auditable trail of thought: the difference between “I just know” and “Here’s how I know.”

VSI replaces the intuition of AI with the discipline of science. It makes intelligence measurable, repeatable, and reproducible across hardware, software, and time. When AI can show its work, it becomes a reliable instrument—not just a “sometimes brilliant” prediction engine.

VSI defines the standard that separates guessing from knowing.

AIA is what makes intelligence smarter.

The second patent I filed, called Adaptive Intelligence Architecture (AIA), is intelligence that learns itself.

It gives AI the ability to adapt—to see from multiple angles, to observe internal and external context, and to refine its understanding over time. Instead of reacting to data based on guessing, AIA lets intelligence reason through change and learn from what happened. It can shift perspective, connect patterns across domains, and evolve its models of the world as new information arrives.

AIA replaces static learning with dynamic cognition. It makes intelligence fluid—capable of teaching itself, questioning itself, and improving itself continuously. When AI can adapt across layers of context, it becomes more than analytical; it becomes intuitive in a way that’s structured, deliberate, and self-directed.

AIA is the standard that separates random regurgitation and memorizing from understanding.

AIN is what keeps it safe for humans.

Another patent filing, called Aligned Intelligence Network (AIN), details how intelligent systems stay consistent, coordinated, and safe as they grow.

It connects multiple intelligences into a shared framework—so they can learn from each other, understand collective goals, and maintain stability even as they evolve. AIN gives AI the ability to communicate, reason, and collaborate while staying anchored to common principles.

AIN replaces isolation with coordination. It makes intelligence interoperable—able to build, debate, and refine together without losing coherence or control. When systems can align with one another in real time, they stop competing for truth and start converging on it.

AIN is the standard that separates powerful intelligence from trustworthy intelligence.

The Backbone of AGI

VSI tracks and logs intelligence so it can be verified and reproduced.

AIA makes intelligence actually smarter every time (not just randomly).

AIN keeps intelligence, even autonomous intelligence, safe and useable for humans.

Together, these inventions form the backbone of artificial general intelligence—the foundation every future system will need to think clearly, learn endlessly, and operate safely.

AGI without this three-layer structure isn’t progress—it’s risk. And it’s worthless since we can’t use it in real life.

With it, intelligence becomes accountable, adaptive, and aligned with human goals.

VSI, AIA, and AIN supply the framework that turns artificial intelligence into something worthy of trust.

They’re the path from prediction to understanding, from automation to awareness.

I didn’t just imagine AGI.

I built the architecture that makes it possible.

⸻

Read more about my research & patent portfolio at tldragi.com

Today, the official ChatGPT account follows exactly one other official ChatGPT account: ChatGPT India.

Not the United Kingdom, not Germany, not Japan — just India.

That small digital gesture captures a larger truth.

OpenAI has made deliberate noise about only one foreign market, and it isn’t coincidence or courtesy. It’s strategy.

When a company of that size starts investing disproportionate attention, building local policy teams, experimenting with pricing, and integrating with domestic payment rails, it’s telling you something about where it can still operate freely — and where it can’t.

Right now, that place is India.

Every other major jurisdiction has begun tightening the perimeter around AI.

In the United States, regulation has become a political sport — senators perform comprehension on camera while the FTC drafts threats it cannot enforce.

In Europe, compliance has metastasized into theology: every dataset must come with provenance, every prompt with provenance of its provenance. China is closed to foreign labs. Japan and South Korea are demographically old and linguistically narrow.

And then there’s India — open, English-speaking, and accelerating.

OpenAI is not expanding there. It’s escaping there.

The One Market That Still Moves Like 2015

Every frontier company eventually hits the same wall: scale collides with law.
In America, that collision has already happened.

Every new capability triggers a congressional hearing, a lawsuit, or a petition to the Copyright Office.

The system that birthed OpenAI now constrains it.

India, by contrast, still operates like the internet used to — messy, fast, and ambitious. It has laws, but not landmines.

It has regulators, but not inquisitors.

It is still possible to build first and argue later.

For a company whose product is perpetual iteration, that difference is existential.

OpenAI can deploy, test, and refine in India at a velocity that would be legally suicidal in California or Brussels.

It’s not about laxity; it’s about philosophy. India still believes that technological growth is inherently national growth.

Most Western economies no longer do.

Demographics as Destiny

India is the world’s youngest major country.

The median age is 28. Nearly a billion people are online, and hundreds of millions speak English.

Data is practically free. Smartphones are ubiquitous. The friction between access and curiosity is almost zero.

Every one of those factors matters for AI adoption.

The U.S. is an aging, saturated market.

Europe’s population is static. Japan’s is shrinking. China’s is large but inaccessible to U.S. firms.

India is the only population at continental scale that is both growing and reachable.

It’s also unusually bilingual in the way that matters for AI.

In India, English is the operating language of education, law, and commerce — but most users think, text, and search in a hybrid of English and local dialects.

That code-switching produces a linguistic diversity that no Western dataset can replicate.

It’s exactly the kind of language behavior a model like ChatGPT learns from best: flexible, unformal, globally comprehensible.

So while the rest of the world debates whether AI will replace jobs, India is producing the kind of usage that trains it.

The Economics of Access

OpenAI’s new India-only subscription tier — ChatGPT Go, priced at about four dollars a month — isn’t charity. It’s calculus.

It’s a controlled experiment in what happens when you price AI for a middle-income country with extreme digital density.

In the U.S., a $20 subscription feels discretionary. In India, $4 is a commitment.

That difference turns every paying user into a long-term behavioral study — what they use, what they abandon, what they teach the system to prioritize.

If the economics work there, they work anywhere.

India gives OpenAI the one thing the American market can’t: elasticity. It lets the company model what AI adoption looks like when it’s no longer a luxury product for professionals but a utility for everyone.

The Payments Superpower

India’s infrastructure advantage goes far beyond population. It’s institutional.

The country’s Unified Payments Interface — UPI — is the most advanced digital payments network in the world. It moves money instantly, freely, and at national scale.

There is nothing comparable in the United States, where the financial system still treats speed as a premium feature.

For OpenAI, that matters because AI’s next frontier isn’t conversation — it’s transaction. The company is already testing integrations that let users buy, book, and pay directly through ChatGPT.

India is the only market where those integrations are technically and legally simple to test.

If you want to know what the economic model of AI will look like in five years, watch what happens when ChatGPT starts using UPI. It’s the first real glimpse of an autonomous assistant that can act, not just advise.

The Talent Mirror

Another reason India functions as OpenAI’s natural extension is that half of the global AI workforce already speaks with its accent.

For decades, American tech companies have trained Indian engineers through their graduate programs, employed them on H-1B visas, and eventually exported them back when immigration policy tightened.

Those engineers now lead, fund, and staff much of the domestic Indian tech ecosystem.

The intellectual pipeline never closed; it reversed.

When OpenAI opens a Delhi office, it’s not entering foreign territory.

It’s building on an existing neural network of people who already built Silicon Valley once.

That feedback loop gives the company an advantage it can’t replicate in Latin America, Eastern Europe, or Southeast Asia.

India isn’t just another engineering hub. It’s the only one fluent in both the technology and the culture that created it.

A Government That Wants to Win, Not Warn

India’s government understands what most others pretend not to: regulating a technology you don’t own is a losing game.

Instead of trying to slow the frontier labs, it has tried to partner with them. It talks about safety but funds acceleration. It courts OpenAI, Microsoft, and Google while simultaneously building its own state-backed AI ecosystem. It issues guidelines, not injunctions.

This creates the rarest political climate in technology: one where government and industry actually share an incentive. Both want to make India central to the global AI economy.

That alignment is fragile, but for now it exists — and for OpenAI, it’s gold. It can test models, run enterprise pilots, and shape policy at the same time. There is no other country where all three are simultaneously possible.

The Cultural Divergence

The West treats AI as a moral problem. India treats it as an economic opportunity.

In the United States, the AI debate is dominated by fear: of job loss, misinformation, bias, liability. In India, it’s dominated by aspiration: productivity, education, entrepreneurship, access. The same technology that makes an American lawyer nervous makes an Indian founder ambitious.

That difference in national psychology is profound. It means Indian society is not building defensive infrastructure around AI; it’s building with it. The country is skipping the phase of moral panic that has paralyzed policy elsewhere.

For OpenAI, that difference is the real prize. It gets a massive, enthusiastic user base that treats AI as a ladder, not a threat. And because that usage happens in English, the data flows directly back into the company’s global models. India is not just a market; it’s a training set that talks back.

Why None of This Can Be Replicated

Other countries offer fragments of the same equation. Brazil has energy and scale but not English. Indonesia has youth but not connectivity. Europe has wealth but not flexibility. The United States has everything except permission.

Only India has all of it at once: scale, speed, English, ambition, and a government willing to absorb risk in exchange for relevance. That’s why it isn’t just another growth market. It’s the last open field for genuine experimentation.

For OpenAI, India isn’t a bet. It’s an inevitability.

The Real Story

OpenAI’s fascination with India is not cultural, sentimental, or even primarily economic. It’s structural.

It’s the one place left where the company can test the next version of the world — socially, legally, and technically — without running into a wall.

India offers what America once did: a population that sees new technology as progress, a government that treats innovation as nation-building, and a market that still measures success by what’s possible rather than what’s safe.

That combination is vanishing everywhere else.

So when people wonder why OpenAI seems so fixated on India, the answer is simple: it’s the only country still moving at the speed that built OpenAI in the first place.

And for a company that survives on acceleration, that’s not fascination. It’s survival.

Bottom Line

American founders, investors, and policymakers should care about India for the same reason OpenAI does: it’s the control group for the future.

India is the only major market still young, fast, and legally flexible enough to run the experiments Western law now forbids.

Whatever works there — payments inside AI, autonomous assistants, large-scale enterprise adoption — will set the blueprint for how those systems eventually reach the United States.

If OpenAI can build and test in India what it can no longer risk at home, then the next generation of AI business models, legal frameworks, and user behavior won’t be born in San Francisco. They’ll be imported from Delhi.

For Americans, that’s the real headline. India isn’t just OpenAI’s favorite market; it’s the proving ground for what will later govern ours.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Obviously, she’s not human.

Her name is Tilly Norwood, and she’s being marketed as the world’s first “AI actress.”

Not a cartoon, not CGI layered over a body, not a deepfake stitched together from stolen faces — a completely synthetic screen personality created by a London studio called Particle 6, through its AI division Xicoia.

Tilly first appeared at the Zurich Film Festival in Switzerland in September 2024, in a short parody film called AI Commissioner.

Most Americans never heard about it because it looked like a European tech curiosity — the kind of small-budget, half-serious demo that comes and goes in a week.

But Zurich wasn’t a stunt.

It was a proof of concept: a public test to see if audiences would accept an actor who doesn’t actually exist.

Fast-forward a year to now, and that quiet demo has become a business pitch. Xicoia has released new high-definition reels of Tilly talking, acting, even doing interviews, and the studio is now shopping her to agencies, brands, and streaming platforms.

All of this is happening at the exact same moment Hollywood is rewriting its union contracts to define what counts as an “AI performer.” The timing is perfect — and deliberate.

What she actually is

Here’s the part that almost everyone gets wrong.

Tilly is not a chatbot with a face, and she’s not an AI filter pasted onto someone real.

She’s built from a combination of generative video, motion, and voice systems that can create her appearance, her speech, and her behavior directly from text or direction.

When her studio writes a script, they don’t cast anyone or rent a camera.

The system literally generates her performance — face, tone, gesture, lighting, everything — in software.

That’s what makes her different from CGI or animation.

CGI still depends on human performance: an actor in a motion-capture suit, a voice actor recording lines, a team of animators painting over every frame.

Deepfakes go the other way — they steal identity by copying a real person’s face or voice without consent.

Tilly sits in the middle.

She doesn’t copy anyone and she doesn’t rely on human labor.

She’s a synthetic original, designed from scratch to look and sound real.

And because she isn’t based on a human, her creators can own and license her image the way Disney owns a character.

This isn’t animation on steroids… it’s something closer to a personality franchise that lives inside software.

Why everyone’s suddenly talking about her

If you’re wondering why a European demo from last year is all over your feed now, it’s because she just left the lab.

Over the last few weeks, Xicoia started releasing new footage and quietly pitching her to U.S. agencies and streaming companies. A few viral clips later and suddenly the prototype no one noticed last year has become the industry’s new moral panic.

But make no mistake — this isn’t about art or technology. It’s about business efficiency.

Studios are still recovering from strikes, shrinking profit margins, and a year of production delays.

Every human actor now comes with agents, contracts, opinions, scheduling conflicts, and rights negotiations.

Tilly fixes all of that. She never needs a trailer, a lawyer, or a day off.

What California’s new law actually says

You may have seen the headlines claiming California just “banned AI actors.” That’s not true.

Last fall, Governor Newsom signed two laws — AB 2602 and AB 1836 — that protect people from being cloned with AI without their permission.

In plain terms, the state said studios can’t use a digital replica of a real person unless that person, or their estate, consents.

Those bills also invalidate contract clauses that would let studios swap an actor for an AI copy in future projects.

But none of that touches fully synthetic performers.

The law stops you from stealing someone’s face.

It doesn’t stop you from creating a new one.

So the irony is that these protections actually make fictional AI characters like Tilly more valuable, because they’re legally clean. They don’t trigger consent rules, and they can be used endlessly without a single rights negotiation.

The reaction

SAG-AFTRA (Screen Actors Guild - American Federation of Television and Radio Artists, which is an American labor union representing roughly 160,000 media professionals) immediately condemned the idea. Emily Blunt called it “soulless.” Whoopi Goldberg said it’s what happens when Hollywood “forgets what art is.”

They’re not wrong to worry — but what they’re really describing isn’t the death of art; it’s the loss of control.

For a hundred years, actors have been the most unpredictable part of the studio system.

They unionized, they demanded residuals, they negotiated likeness rights, and they started to own their own IP.

Taylor Swift re-recorded her masters.

Influencers are now hiring lawyers to remove “in perpetuity” clauses from brand contracts.

Creators have finally learned to understand their value.

And when humans start insisting on ownership, corporations start building things they can own completely.

That’s what Tilly is: not a creative experiment, but a contract solution.

Who wins and who loses

The winners here are obvious — studios, advertisers, and investors who get to own a performer instead of hiring one.

They can cast Tilly in a film, a perfume ad, and a political PSA on the same day, in different countries, in different languages, without negotiating a single contract.

The losers are equally clear — actors, agents, and everyone else whose power depended on human scarcity. When anyone can generate a face that never ages or argues, scarcity disappears, and with it, the leverage that built Hollywood in the first place.

But the bigger loss isn’t labor. It’s authorship.

When a performance is generated by code, who counts as the artist?

The person who wrote the prompt, the engineer who built the system, or the algorithm that rendered the emotion?

No one knows yet — and the law hasn’t caught up.

The new Hollywood

Tilly Norwood isn’t replacing actors; she’s replacing the system that actors forced to evolve.

She turns performance into software and software into property. The new California laws protect real people from being copied, but they also opened a path for studios to build performers who never need permission in the first place.

And that’s the real story. This isn’t the end of acting — it’s the beginning of synthetic celebrity, a version of Hollywood where studios don’t have to cast humans to sell humanity.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

This isn’t another AI demo.

It’s one of the first large-scale consumer products to encode copyright-adjacent constraints directly into generation and distribution. Filters, watermarking, and provenance aren’t afterthoughts—they’re the product.

Sora blocks “President Trump,” but generates a similar “President Tuna.”

It refuses real celebrities yet mirrors their cadence.

It stamps every frame with metadata, then tells users they “own” videos the law does not necessarily recognize as protectable.

That contradiction isn’t a glitch.

It’s the business model.

The First Product That Practices Law

Earlier platforms hosted content and argued about ownership later.

Sora bakes the argument into its code.

Its framework:

1. Opt-out, not opt-in. Rightsholders must tell OpenAI not to use their work. Silence equals permission.

2. Consent as a feature. “Cameos” let users upload their likeness so others can remix it within platform limits.

3. Provenance by default. Watermarks and C2PA metadata trace origin and edits. They’re authenticity tools, not legal judgments, but they shift enforcement work to users.

It looks like safety.

It’s really a private copyright workflow hiding in UX design.

Filters That Catch Names, Not Vibes

Sora’s moderation filters block explicit identifiers—names, faces, logos—but still allow stylistic imitation.

A blocked “President Trump” becomes a freely generated “President Tuna.”

A “football player in a Dallas uniform” slides through, colors and stripes intact.

Because the system reads text, not context.

It detects nouns, not nuance.

It guards against trademark use, not trade-dress confusion. Trade dress turns on overall look and feel—something a keyword filter can’t see.

Nominative references can be lawful under fair-use doctrine, but false endorsement and look-alike design still invite risk.

For famous marks, dilution can attach even without confusion.

So most public figures and brands are one euphemism away from a confusion or false-endorsement claim.

This isn’t a safety net.

It’s plausible deniability in code.

The Legal Snapshot

Under U.S. Copyright Office guidance, works lacking sufficient human authorship—selection, arrangement, or editing—are not copyrightable. The Anthropic ruling in June 2025 confirmed that training on lawful copies is fair use; it left outputs for future courts.

State right-of-publicity laws cover name, image, and voice, but not style. When style creates marketplace confusion, plaintiffs fall back on the Lanham Act’s false-endorsement clause or state unfair-competition law.

Trademark and trade-dress cases hinge on consumer confusion, though nominative fair use can shield some reference. Dilution protects famous marks even without confusion.

And under DMCA § 1202, removing or falsifying copyright-management information—like C2PA data—with knowledge that it will facilitate or conceal infringement can trigger separate liability.

Each doctrine points to the same tension: users may have permission to create, but not protection once they do.

The Legal Paradox

The law says one thing.

The app implies another.

Copyright rejects pure machine authorship, yet Sora markets authorship as a feature.

Its terms promise ownership “to the extent permitted by law.”

If a clip copies a protected voice, logo, or scene, liability sits with the user, not the platform.

Users don’t own the asset.

They own the risk.

Counterpoints

Model providers argue that provenance limits harm, regurgitation is rare, filters improve with iteration, and opt-out mechanisms are the only practical route.

But harm lives in downstream confusion, not dataset exposure.

Low frequency doesn’t erase liability for confusing or infringing outputs.

Iteration is not accountability when imitation remains trivial.

And opt-out shifts cost and policing to creators.

The Real Play

OpenAI isn’t clarifying copyright—it’s monetizing its gray zones.

It transfers the burden of policing to rights holders, trains the public to equate compliance screens with legality, and establishes de facto precedent by volume. Billions of generated clips under identical terms become the new normal long before lawmakers respond.

That’s how code rewrites law: through repetition, not argument.

The Loopholes Are the Feature

Those near-misses—the parody names, the unflagged uniforms—aren’t bugs.

They’re calibrated friction points that keep Sora expressive while defensible.

Too strict, creativity collapses.

Too loose, regulators react.

OpenAI’s moderation walks that tightrope intentionally.

It’s not ethics.

It’s risk engineering.

Quantifying the Exposure

Output-based claims span takedowns to statutory damages—up to $150,000 per work for willful copyright, plus possible DMCA § 1202 penalties, and injunctions for false endorsement or trade-dress confusion.

Magnitude varies by case but is material at platform scale.

Governance by Interface

Every default toggle and confirmation box moves liability from the company to the user.

Click “I confirm I have rights,” and the burden shifts.

This is law transformed into product design.

What to Do Now

Founders and product teams should implement prompt-use policies, retain provenance data, flag marks and uniforms automatically, and pre-clear licensed content.

Creators and agencies should assume no copyright in purely AI-generated material, keep watermarks intact, avoid identifiable brands or people, log prompts and outputs, and register human-edited compilations where creative control is real.

Brands and rights holders should file opt-outs, monitor for trade-dress imitation, create licensed AI channels, and prepare DMCA takedown templates that reference C2PA hashes.

Why This Moment Matters

Before Sora, copyright disputes played out in court.

Now they unfold inside interfaces.

Every generation refines social expectations of what “allowed” looks like, turning product defaults into precedent.

Governance has migrated from statute to software.

The Next Lawsuit

Training on lawful data is settled as fair use; outputs are the next frontier.

Expect claims over confusion, endorsement, and removal or falsification of metadata rather than ingestion.

Even if courts reaffirm fair use for training, output liability remains a design problem, not an academic one.

The Bottom Line

Sora didn’t fix copyright.

It privatized it.

The first AI that can make movies also writes law—silently, line by line, inside its prompt box.

For anyone building or investing in generative systems:

Inputs were yesterday’s fight.

Outputs are tomorrow’s liability.

The law hasn’t caught up, but the product already has.

Nothing here is legal advice; risk always depends on facts, jurisdiction, and use.

You don’t need a court ruling to see what changed.

Sora turned copyright into software.

The next copyright office isn’t in Washington.

It’s running in the cloud.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

In 2020, Trump threatened to ban the app outright.

In 2024, Congress passed the Protecting Americans from Foreign Adversary Controlled Applications Act (PAFACA) — a divest-or-ban ultimatum aimed squarely at TikTok’s Chinese parent, ByteDance.

The vote wasn’t close, and it wasn’t partisan.

• PAFACA passed with overwhelming majorities in both the House (on March 13, 2024) and the Senate (on April 23, 2024).

• It was signed into law by President Biden on April 24, 2024.

And since April 24, 2024, TikTok influencers, mainstream outlets, and social platforms have been reporting that “the TikTok ban is actually happening” or “the TikTok sale is on/off.”

This month (September 2025), the standoff ended: Oracle and a group of U.S. investors bought control of TikTok’s U.S. operations for $14 billion.

The headlines said “Trump saves TikTok” or “national security deal.”

That’s not what just happened.

Why TikTok Was Targeted

TikTok is not just another app. It was forced into this sale because of three risks no other major platform combines:

1. Privacy overreach – TikTok collects location data, browsing history, and even how you type when using its in-app browser. That looks uncomfortably close to keylogging (aka screen recording you as you type ANYTHING on your phone - not just on the TikTok app).

2. Foreign leverage – ByteDance is a Chinese company. Under Chinese law, it can be compelled to share data with the state. That possibility alone was enough to make Washington nervous.

3. Algorithmic influence – TikTok’s “For You” feed is the most powerful content engine in the world. It decides what 170 million Americans see every day. That means it can invisibly amplify or suppress narratives.

Facebook tracks you. YouTube addicts you.

But only TikTok added the risk of direct foreign government leverage.

That’s why Washington forced a sale.

The Reach Oracle Just Captured

Here’s the scale Oracle now controls:

• 170 million Americans use TikTok.

• The average user spends about 58 minutes per day on it.

• 57% of U.S. teens are daily users, with especially heavy use among Black and Hispanic teens.

This isn’t just “social media.” It’s prime-time television, radio, the town square, and the high school lunch table rolled into one feed.

The Price Tag: $14 Billion

Oracle and its partners (Silver Lake, MGX) paid $14 billion for TikTok’s U.S. operations. ByteDance keeps a minority stake.

$14B sounds big. But in media terms, it’s cheap. Oracle didn’t just buy an app. It bought the most concentrated stream of youth attention in America.

What That Influence Would Cost the Old Way

Let’s translate Oracle’s deal into advertising math.

• TikTok reaches 170 million Americans daily.

• To show five ads per day to each of them, you’d need 850 million ads every day.

• That’s 25 billion ads per month.

• On TikTok, showing ads at that scale costs about $150 million per month. Multiply by 12 months, and you’re at nearly $2 billion every year.

• Double it to ten ads per person per day, and you’re spending over $3.5 billion annually.

That’s the rental model. Oracle didn’t rent. It bought the building.

And to put that in perspective: the most expensive ad stage in America is the Super Bowl.

• One 30-second commercial costs about $8 million.

• It reaches about 120 million viewers — but only once.

TikTok reaches a Super Bowl-sized audience every single day, and holds them for nearly an hour.

That’s the difference between paying for one stadium shout and owning the stadium itself.

TikTok vs. the “Serious Media”

Now compare TikTok to the institutions that still define “mainstream media”:

• Fox News primetime: about 2 million nightly viewers, for about an hour each → 120 million minutes of attention per day.

• New York Times online: about 90 million monthly visitors, but only 2 minutes per visit → 6 million minutes per day.

• All top 50 U.S. newspapers combined: tens of millions of minutes per day.

TikTok: nearly 10 billion minutes of attention every day in the U.S.

That’s two orders of magnitude more attention than any single outlet. And it skews young.

Who Owns “The Media” (and Now, TikTok)

American media ownership is concentrated in dynasties and conglomerates:

• Murdoch family – Fox News, Wall Street Journal

• Warner Bros. Discovery – CNN

• Comcast – NBC, MSNBC

• Disney – ABC, ESPN

• Paramount (Redstone family) – CBS

• Sulzberger family – New York Times

These are the institutions that have defined “media power” for decades.

Now add Larry Ellison. Except he didn’t buy a cable network or a newspaper. He bought the feed that makes all of them look small.

And here’s another connection: most of these conglomerates share the same large institutional investors — Vanguard, BlackRock, State Street. They don’t coordinate editorial, but capital ties them together. Oracle now joins that club, with a far more powerful property.

The Other Oracle Moves That Matter

This isn’t Ellison’s only power grab. At the very moment he was buying TikTok, Oracle was also striking some of the biggest infrastructure deals in tech history:

• A $30 billion annual deal to provide cloud capacity to OpenAI.

• A partnership with OpenAI and SoftBank to build five new AI data centers under the $500 billion “Stargate” project.

• An $18 billion bond raise to fund data center expansion.

• A leadership shift, with Oracle splitting the CEO role into two to emphasize cloud and AI.

Ellison isn’t just buying the screen time — he’s buying the servers behind it.

And the impact is visible. Oracle’s stock surge on these moves pushed Larry Ellison past Elon Musk to become the richest person on earth. The symbolism matters: Musk spent $44B to buy Twitter/X and make himself a media power. Ellison just spent $14B to buy TikTok’s U.S. feed — and got more daily attention for one-third the price.

So Oracle didn’t just buy a media empire. It’s also building the compute backbone for the AI systems that shape what people see, hear, and read next.

Media + compute is not coincidence. It’s strategy.

Was $14B Cheap or Expensive?

At today’s ad prices, TikTok’s U.S. reach would cost $2–4 billion a year to rent. Oracle bought it for about the cost of four to eight years of ad spend. After that, it’s essentially free.

For comparison: $14B might buy you a struggling newspaper chain or a fading cable network. Those audiences are older, shrinking, and fragmented. TikTok’s audience is young, growing, and locked in for nearly an hour every day.

Ellison didn’t just buy an app. He bought more daily attention than Fox, CNN, NBC, and the New York Times combined — for less than Paramount is worth.

🧠 Bottom Line

The headlines say “Trump saves TikTok.”

The truth is simpler: Oracle just bought America’s attention span at a clearance price.

For $14B, Ellison purchased more daily influence than Fox, CNN, NBC, and the New York Times combined. At advertising rates, that much reach would cost billions every single year to rent. Now Oracle owns it outright.

It’s a reminder that influence isn’t abstract. It has a price tag. It can be bought.

Next time you open TikTok, remember: you’re not just scrolling. You’re inside the cheapest $14 billion deal for influence in history.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

• Nvidia just promised to “invest” up to $100 billion in OpenAI. That’s not a typo. One hundred. Billion. Dollars.

• OpenAI will then spend that money buying Nvidia’s GPUs. Nvidia records the sales as revenue while also holding an equity stake in OpenAI.

• Because OpenAI isn’t profitable, all that spending turns into future tax shields. They can roll those losses forward for years, wiping out taxes when profits finally arrive.

What’s happening

OpenAI says it needs 10 gigawatts of compute to train its next generation of models. That’s a number so big it’s easier to compare to national power plants than to cloud servers. It works out to hundreds of billions of dollars of data centers, chips, and energy.

No startup—no matter how hot—can raise $100 billion in cash. Even with Microsoft in its corner, OpenAI can’t go to the market and say, “we’d like $100 billion please.”

So Sam Altman asked the one person who could solve the problem: Jensen Huang, the CEO of Nvidia, the only company that makes the GPUs OpenAI needs.

The result: Nvidia “invests” $100 billion into OpenAI. In reality, it’s staged. As OpenAI builds each chunk of its new compute infrastructure, Nvidia puts money in, and OpenAI uses it to buy Nvidia hardware.

On Nvidia’s books: $100 billion in new sales. On OpenAI’s books: $100 billion in assets (GPUs and servers) plus a mountain of tax-deductible depreciation. Everyone wins on paper.

How we got here

• Early September 2025: OpenAI unveils “Stargate,” a multi-site U.S. data center project, aiming for several gigawatts of power and $400B+ of eventual build costs.

• September 22, 2025: Nvidia and OpenAI sign a letter of intent: up to $100B in staged “investment,” tied to 10 GW of Nvidia systems. First 1 GW goes live in late 2026.

• September 23, 2025: Regulators and analysts immediately flag antitrust questions: can the dominant GPU supplier bankroll its star customer without squeezing everyone else?

Why this structure works (for them)

If OpenAI had tried to raise $100 billion the usual way, here’s what it would look like:

• Equity raise: At a $100B valuation, that’s 50% dilution. No investor group would tolerate it.

• Debt: At 8% interest, $100B in loans means $8B a year in payments—unworkable with no profit stream.

• Chip diversification: Splitting orders between Nvidia, AMD, or Intel would require rebuilding OpenAI’s training software stack mid-race. Too slow.

The vendor-financing loop solves all three problems at once: no dilution, no back-breaking debt, and instant continuity on Nvidia hardware.

The hidden accounting trick

Here’s what no press release says: OpenAI’s current unprofitability makes this structure even more powerful.

• All those GPUs are capital assets. They get depreciated over 3–5 years.

• That depreciation shows up as losses. But since OpenAI is already losing money, it just adds to the pile of net operating losses.

• Those losses can be rolled forward for years. So when OpenAI does turn profitable, it can apply those losses against future income—meaning it could book billions in profits and pay almost nothing in taxes for a long stretch.

Amazon, Tesla, Uber—this is exactly how they played their early years. Scale fast, pile up losses, then let the tax code do the cleanup later.

The math in simple terms

• Nvidia “invests” $25B.

• OpenAI spends $25B buying GPUs.

• Nvidia books $25B in revenue and still holds $25B in OpenAI equity.

• OpenAI books $25B of servers and GPUs as assets. Depreciates ~$5–8B of that each year, building future tax shields.

Repeat 4 x, and you have a $100B cycle.

For Nvidia, it’s double-dipping: revenue today plus equity upside tomorrow. If OpenAI’s valuation doubles, Nvidia’s $100B stake doubles too—on top of the $100B in product sales it already booked.

Why it matters for business leaders

• This isn’t a loophole. It’s how Silicon Valley has always done scale: align your supplier’s incentives with your financing gap, and suddenly the impossible number ($100B) becomes doable.

• The optics matter. To Wall Street, Nvidia can say: “We invested in AI’s future.” To regulators, OpenAI can say: “We secured compute without creating debt risk.” Both technically true, but the real story is the loop.

• The tax shield is intentional. Losing money now is part of the plan. Those losses will wipe away future tax bills when revenue finally hits.

• Copycat alert. Any business facing massive CapEx—energy, chips, space, infrastructure—should study this model. Supplier as investor isn’t new, but $100B makes it the template for the next decade.

What’s next

• Regulators will comb through the deal to see if Nvidia has to make supply non-discriminatory.

• OpenAI’s real bottleneck isn’t chips—it’s power. Stargate sites need to get connected to the grid, or the whole plan slips.

• Watch for leaks about redemption rights or buyback clauses. Deals this size always have hidden protections.

The takeaways

• $100 billion isn’t a typo. It’s the size of the loop, and it only works because the supplier is also the investor.

• Circular but brilliant. Money goes in, money comes back, everyone’s numbers look better.

• Depreciation is a feature. Losses today become tax shields tomorrow.

• This is the new playbook. If you’re building at scale, don’t just ask for capital. Ask your suppliers to finance your growth.

This is how Silicon Valley does deals: bend the structure, optimize the optics, lock in the growth, and let the tax code take care of the rest.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

• What is H-1B? A U.S. work visa for highly skilled jobs (software engineers, chip designers, researchers).

• What just changed? For any new H-1B filed on or after Sept 21, 2025, the government now requires a one-time $100,000 payment with that petition. Renewals of existing visas are not charged.

• How long does this last? The order runs 12 months unless extended.

• Can anyone skip the $100k? Yes. DHS can grant a “national-interest” waiver for certain people, companies, or industries.

What just happened

On Sept 19, the White House issued a proclamation: every new H-1B petition must include a $100,000 payment. This isn’t a filing fee — it’s a toll on entry.

Markets reacted immediately. Indian IT stocks fell. Several big employers told H-1B staff to avoid non-essential travel. India’s foreign ministry warned of family disruption if cases get delayed.

Founder voices came back into focus. Elon Musk has said bluntly that H-1B is why he and many of the people who built modern tech could work in the U.S. He once vowed to “go to war” over restricting it. Translation: tech leaders see this as an attack on the talent pipeline, not just a paperwork tweak.

The political rationale

The administration’s message is blunt: companies kept bringing in H-1B workers while laying off Americans.

The White House fact sheet led with a killer number: H-1B holders now make up 65% of the U.S. IT workforce, up from 32% in 2003.

Then came the examples:

• One company laid off 27,000 U.S. employees since 2022 while receiving 25,075 new H-1Bs.

• Another got 5,189 H-1Bs in FY2025 while cutting 16,000 U.S. jobs.

• A third received 1,698 H-1Bs while announcing 2,400 layoffs in Oregon.

The punchline: unemployment among recent U.S. computer-science grads rose from 6.1% to 7.5% while companies kept filing H-1Bs.

Plain English: the White House is pitching the $100k toll as a reset button. Hire globally if you must, but pay a premium. If not, hire locally.

The legal and strategic angle

• Can a president do this? Yes. Presidents have broad power to restrict who can enter the U.S.

• What’s unusual? Attaching a $100,000 payment. Normally, fees are set through rulemaking. Expect lawsuits arguing this “fee” is not authorized that way.

• Business reality: Until a court says otherwise, assume the toll applies to every new H-1B, unless a waiver is granted.

The math that matters

• A typical U.S. software developer earns about $133,000. With benefits, call it $166,000 a year.

• Add the toll, spread over three years, and the cost rises to about $199,000 a year.

• That’s an extra $33,000 per year, or roughly a 20% increase.

At company scale:

• Mega-cap example: 3,500 new H-1Bs = $350 million in tolls. If half are waived, that’s $175 million. For firms earning $60–110 billion, this is about half of one percent of profits — affordable, but not for every role.

• Mid-size software company: 20 new H-1Bs = $2 million. On $40 million in profit, that’s 5% gone.

• IT services provider: 1,000 new H-1Bs = $100 million. On $1 billion in profit, that’s 10% wiped out. Expect offshore hiring to accelerate.

Startups: $2 million in tolls can equal six months of runway. Most will only pay for one irreplaceable hire and move everything else to other visas or distributed teams abroad.

The waiver wildcard

The proclamation lets DHS grant national-interest waivers for people, companies, or industries.

• If AI, semiconductors, and defense get waivers, the toll is a scalpel.

• If not, companies will shift roles to Canada, Mexico, or India.

Action item: For each new role, run three lanes:

1. Pay/waive for mission-critical.

2. Alternate visa (O-1, L-1, TN, E-3).

3. Relocate offshore or near-shore.

Global competition for the same talent

If the U.S. slows H-1B entry, other countries are ready:

• Canada: Indians got nearly half of all skilled-migration slots in 2023; over half in STEM draws.

• European Union (Germany): Indians received about one-quarter of EU Blue Cards; Germany issued most.

• United Kingdom: Indians remain the largest nationality on Skilled Worker visas, even after a clampdown.

• Australia: India led skilled migration in 2022–23 with over 36,000 slots.

• UAE: Dubai issued about 158,000 Golden Visas in 2023, many targeting professionals.

Plain English: If America taxes new H-1Bs, Canada, Germany, the UK, Australia, and the UAE will happily take the same engineers and founders. The work gets done — just not here.

Follow the money

Big Tech vs. startups

• For a mega-cap, $350 million is a few days of revenue. They’ll pay for frontier hires and reroute the rest.

• For a startup, $2 million is six months of life. They’ll avoid the toll except for a single company-changing hire.

Runway reality

• A startup with $20 million raised and a $3 million monthly burn has 7 months of runway. Add $2 million in tolls and that drops to 6 months — with no acceleration in output.

Local wages vs. the toll

• If fewer H-1Bs arrive in hubs like Seattle or the Bay Area, salaries rise. A 10% bump on a $166,000 loaded cost = $16,000 more per engineer per year. For 1,000 engineers, that’s $16 million — about half the cost of paying the toll for 500 H-1Bs.

Capital tradeoffs

• Every dollar spent on tolls is a dollar not spent on R&D, hiring, or launches. For giants, it’s noise. For mid-size firms, it cancels a product. For IT services, it drives margins into single digits.

The bigger business picture

This won’t stop the AI boom. It will decide where the boom happens.

• Big Tech will pay selectively.

• Startups will default distributed.

• Indian IT loses margin now but gains long-term talent density.

• U.S. competitiveness depends on waiver breadth and whether America still wants to be the default home for global talent.

What to watch

• Court challenges over whether a president can attach a $100k “fee.”

• DHS waiver criteria — narrow vs. broad carve-outs.

• Company earnings calls quantifying immigration costs.

• Visa substitution: more O-1, L-1, TN, E-3; fewer H-1Bs.

• Talent flows: more Indian engineers choosing Canada, Germany, the UK, Australia, UAE.

The takeaway

This is a talent tax. It adds about $33,000 per year to each new H-1B hire’s cost.

• Pay the toll for mission-critical roles.

• Reroute when another visa fits.

• Relocate when neither works.

Politically, the White House frames this as protecting U.S. jobs. Founders counter that H-1B built modern tech. Both can be true.

The difference is strategy. If waivers are broad, this toll is noise. If not, the U.S. is sending the next wave of AI builders a clear message: build here if you can afford the tax — or build the future somewhere else.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

This is the first mirror we’ve had on human behavior with AI, and the reflection is not what you’d expect.

What the Study Looked At (and What It Didn’t)

• It only measured prompts. What people typed. Not how good the answers were.

• It only covered consumer plans. Free, Plus, Pro. No Enterprise or Education accounts, no logged-out users.

• It sorted prompts into categories. Writing, Coding, Practical Guidance. And whether users were mostly asking for advice or asking for finished work.

Think of it as a usage map, not a performance review. It tells us where people are steering the model, not whether it drove well.

Work Is Losing Ground to Personal

Non-work use rose to 73%. Work dropped to 27% in mid-2025.

At work, “Writing” is the top task—but mostly editing. Two-thirds of prompts are polishing, translating, or summarizing text humans already wrote. Less “write my draft,” more “fix my draft.”

Personal use is exploding. People lean on ChatGPT as a daily tutor, translator, and explainer.

So the consumer app is becoming more personal than professional. The heavier-duty work is shifting into enterprise accounts, coding copilots, and integrations that this study didn’t track.

Why Coding Looks Small

Coding shows up as ~4% of prompts. That doesn’t mean developers abandoned AI. It means they left the consumer chat window.

This matters because “AI for coding” has been one of the loudest headlines. Social media and tech press have painted coding as the killer use case—copilots writing apps, AI replacing junior developers. So it’s striking that in this dataset, coding looks small.

Serious work happens in:

• Copilots inside code editors like VS Code.

• Automated agents that run sequences of tasks.

• APIs (digital doorways) that connect ChatGPT directly to business systems.

That traffic doesn’t show here. So the 4% figure is about one interface, not the whole reality.

Who’s Driving Usage: Age Matters

Almost half of all prompts came from users under 26:

• Younger users blur personal and work, mix homework with hobbies, and experiment more.

• Older users log fewer prompts, and keep work and personal separate.

The charts tilt young—and that shapes what we see.

Is Work Undercounted?

Yes, but only partly. The study excluded whole buckets of workplace use. OpenAI says 80% of the Fortune 500 have ChatGPT accounts tied to corporate emails, but only ~5% have true Enterprise contracts. Global business users are around 600,000—tiny compared to ~33 million U.S. businesses. So enterprise use exists, but penetration is still single-digit.

Logged-out prompts are missing too, and the classification system sometimes blurs “personal” vs. “work.”

Context helps: McKinsey (2025) found almost all large firms report some generative AI use, but only ~1% call themselves “mature.” BCG (2025) found only 36% of employees feel trained; even 5–10 hours of training multiplies adoption.

So yes, the study undercounts work—but hidden enterprise licenses don’t secretly outweigh the personal tilt. Many businesses still rely on personal accounts for work, which means plenty of real work still shows up in the consumer logs.

Asking vs. Doing

People use ChatGPT more for asking than doing. By mid-2025, consumer prompts split: 52% Asking, 35% Doing, 13% Expressing.

At work, “Doing” rises to 56%, but a third of those are Writing→Editing. Even at work, ChatGPT is often an editor, not a first drafter.

This shows most people see it as a tutor, explainer, or thought partner—not yet as a worker producing finished products.

Prompt Length: Why It Matters

User’s prompts are short. Almost all under 250 characters. Most under 50.

That’s a problem. Short prompts are vague prompts. Most people aren’t giving ChatGPT enough to work with—which is why most outputs feel shallow.

The longer and clearer the instructions, the more powerful the output. Yet most users are keeping prompts short—and leaving capability on the table.

Why This Matters

Most news write-ups will stick to: “Personal dominates. Writing is top. Coding is small.” All technically true—but surface-level.

The deeper reality:

• Consumer usage isn’t the whole story. Enterprise use exists but is still thin compared to the consumer wave.

• Work writing is mostly editing. Productivity won’t jump until more people let AI do the first draft.

• Coding isn’t disappearing—it moved.

• Prompts are short. Most users aren’t giving AI the clarity it needs or using it to its full potential.

• Training is weak. Companies turned it on, but didn’t teach people to use it.

Why You Should Care

This isn’t trivia. It’s a roadmap of adoption. If most people keep prompts short and shallow, you gain an edge by teaching your team to go deeper. If work is shifting into enterprise tools, early adopters of training and integration will outpace dabblers. And as personal use explodes, customers will expect instant AI help as the norm—ready or not.

Bottom line: OpenAI’s new study is a mirror on how people actually use ChatGPT. It shows consumer users lean personal and young. The bigger workplace story is happening offstage—in enterprise accounts, copilots, and integrations. Until people learn to give longer, clearer prompts, they’ll keep using a jet engine like it’s a bicycle bell.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Instead of clicking links — maybe yours — they get a tidy AI answer at the top of the page. They read it, move on, and you never see them.

Your Customers Might Never Reach Your Site Again

That’s happening now.

And Rolling Stone is suing Google over it.

(Yes, yet another media company suing yet another Big Tech AI company.)

Rolling Stone Took Google to Court — But It’s Really About You

In September 2025, Rolling Stone’s parent company sued Google.

Their accusation: Google turned from a search engine into an answer machine that steals the clicks businesses rely on.

They point to three things:

1️⃣ Traffic collapse. 20–34% fewer clicks when AI summaries appear. Affiliate revenue already down by a third.

2️⃣ Training without consent. Millions of Rolling Stone articles allegedly fed into Google’s Gemini AI for free.

3️⃣ Abuse of monopoly power. Publishers can’t say “no” without disappearing from search entirely.

The case headlines with Rolling Stone, but the subtext is clear: any business that counts on Google traffic to its site is at risk.

And meanwhile, Google isn’t just building features — it’s winning eyeballs.

Gemini just hit #1 in the U.S. App Store for the first time.

That means more users installing the app, generating outputs, and consuming information inside Google’s ecosystem rather than clicking out to the open web.

How This Ends: Google Wins (Almost) Everything

Here’s my contrarian prediction… Rolling Stone’s lawsuit makes headlines, but in court Google will walk away with almost everything intact.

Training is fair use. Judges have already said using lawfully obtained text to train AI is “highly transformative.” Rolling Stone’s “you trained on us” argument is weak (and Rolling Stone makes no claim that Google’s acquisition of articles was unlawful).

Outputs aren’t copy-paste. Google’s AI summarizes, it doesn’t regurgitate. That’s annoying for publishers but legally safer.

Antitrust courts don’t redesign products. Expect token remedies: attribution tweaks, maybe opt-out knobs, some licensing for optics. Not a dismantling of AI Overviews.

Opt-out exists, even if painful. Google can point to robots.txt and nosnippet. Courts see trade-offs as business choices, not coercion.

Traffic was never yours. Free clicks from Google were rented, not owned. The landlord just raised the rent.

The likeliest outcome? Google keeps the AI Overviews. Rolling Stone gets little more than PR mileage from the suit. And the rest of us are left to adapt to an answer-first internet whether we like it or not.

The “Regurgitation” Accusation — Explained Like a Human

Publishers say Google’s AI “regurgitates” (“spits out an exact copy of”) their work. That sounds like Google is copy-pasting whole articles.

That’s not what’s happening.

AI models don’t regurgitate, they generate. They learn patterns and then predict the next word — like autocomplete on steroids.

So what shows up in AI Overviews is usually a summary or paraphrase, not a literal copy.

But here’s the sting: for the user, it doesn’t matter. A two-sentence summary still replaces the click. If Google tells them “the top three wireless earbuds are X, Y, Z,” your detailed review is irrelevant — even if it wasn’t copied word-for-word.

Why This Fight Should Terrify Every Business With a Website

Rolling Stone’s complaint could have been written by any company that depends on traffic from searches:

• The old deal: You write content, Google scans it, users click through to you.

• The new deal: You write content, Google scans it, Google answers directly, you never see the user.

If the court forces Google to pay or limit this, companies get breathing room.

If not? Google keeps the eyeballs, and your website becomes optional.

Winners, Losers, and the Ones Who Don’t Know They’re Losing Yet

Winners: Google (ads), users (faster answers), advertisers (more inventory).

Losers: Rolling Stone today. Your company tomorrow.

Because this isn’t about journalism. It’s about whether your landing pages, blogs, or product guides ever get seen again.

How to Make Sure Your Website Still Matters

Become the Source AI Quotes: Publish data, frameworks, and takeaways that AI will lift and cite.

Put Value Beyond the Summary: Offer what a one-paragraph Google answer can’t: tools, calculators, proprietary insights.

Build Channels Google Can’t Steal: Email lists, podcasts, social media followings, communities. Direct distribution is your insurance policy.

Stop Treating SEO as a Free Lunch: Search clicks were always rented, never owned. Treat them as one channel, not your foundation.

The Harsh Truth You Can’t Ignore

Rolling Stone’s lawsuit may drag on for years. But the economics already shifted.

If your growth depends on free traffic from Google, you’re living on borrowed time.

Because Google didn’t just kill SEO for publishers. It killed SEO for everyone. And unless you change course, it might make your website (and your business) irrelevant too.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

It’s in the middle of transforming into a Public Benefit Corporation — a Delaware legal form where directors have to balance profit with a stated mission — while also renegotiating its long-standing deal with Microsoft.

The reason is simple: ChatGPT may be the most popular software product in the world, but the economics don’t add up.

• Adoption: ~700–800 million weekly users.

• Revenue: About $2 billion annually in late 2023 → roughly $12 billion by mid-2025 if today’s pace continues. That’s just $1–1.50 per user per month.

• Costs: Analysts estimate annual infrastructure costs at $20–40 billion (GPUs, energy, and cloud hosting). That means OpenAI is effectively spending $2–3 for every $1 it earns.

• Subsidy: Microsoft’s billions and other outside capital are covering the gap, alongside the small minority of paying users who cushion the blow.

Which means the cheap, abundant AI we’ve all been enjoying isn’t sustainable. It’s a subsidy phase. Once AI is fully woven into our daily work and lives, OpenAI will have no choice but to raise prices and shrink the free tier. By then, walking away won’t be realistic.

How we got here

• 2015 — Nonprofit launch. OpenAI began as a nonprofit lab, backed by $1 billion in pledges. Mission credibility, but financially handcuffed.

• 2019 — Microsoft lifeline. Created a capped-profit company. Microsoft invested $1 billion and got exclusive hosting rights on Azure. Solved infrastructure costs; created governance headaches.

• 2022 — ChatGPT explosion. Launched November 30. 1 million users in five days. By 2025: ~750 million weekly users. Costs skyrocketed.

• 2023 — GPT-4 and board drama. Revenue if today’s pace continues: ~$2 billion. In November, Sam Altman was fired and rehired within five days — exposing the fragile governance model.

• 2024 — Bigger models, bigger bills. OpenAI previewed video generation, dramatically more expensive to run. Revenue if today’s pace continues rose to $5.5 billion by year-end.

• 2025 — Restructure or stall. Revenue if today’s pace continues climbed to $10–12 billion, but infrastructure costs were estimated at $20–40 billion per year — several times higher than revenue. OpenAI announced its Public Benefit Corporation conversion and signed a memorandum of understanding with Microsoft to loosen legacy deal terms.

Today’s Bargain, Tomorrow’s Bill

OpenAI’s reported revenue has grown quickly — from about $2 billion a year in late 2023 to roughly $12 billion a year by mid-2025 if current usage levels continue. On the surface that sounds huge. But spread across ~750 million weekly users, it works out to only about $16 per user per year — because most people aren’t paying anything at all.

The real question is where that $12 billion actually comes from. Here’s how it breaks down:

• Almost everyone is free. ~98% of ChatGPT users don’t pay anything.

• Only a tiny fraction pay. ~2% have a Plus plan at $20/month ($240/year). That’s ~10 million people, which adds up to about $2.4 billion annually in subscription revenue. Add in some business deals and Microsoft’s support, and you get to the ~$12 billion figure.

• The imbalance. Most users pay $0, a small minority are paying $240+ each year, and outside money is still doing the heavy lifting.

Even at a personal or micro/small-business level, this bites… I run a company with fewer than 20 employees, and I’m already spending thousands of dollars each month on ChatGPT Pro. At $200 / month for each seat, that’s real money — and I’m just a tiny drop in the user base. If businesses like mine are paying that much, while the overwhelming majority of users pay nothing, you can see how distorted the economics are.

The bigger picture is unavoidable: serving 100s of millions of free users costs an estimated $20–40B per year. Against $12B in revenue, the model bleeds money. That’s why the free ride cannot last.

This raises some uncomfortable questions:

1. Do paying users end up footing an even larger bill? As costs grow, will the subscription price climb to $30–$40 per month, essentially making the minority of paying users subsidize hundreds of millions of free riders?

2. Or does the free tier disappear? OpenAI could eventually cut or sharply restrict free access, forcing everyone to pay at least something to offset infrastructure costs.

3. And the broader issue: if free access disappears entirely, does that create an access-to-justice problem — a world where only those who can pay get to participate in the new baseline of work and knowledge?

Right now, our cheap (or free) ChatGPT subscriptions are being underwritten by Microsoft’s billions and investors’ patience.

Just like Uber rides were once subsidized by venture dollars.

And just like Uber, the plan is to raise prices once we can’t imagine life without it.

Why you should be concerned now

• Dependency is the strategy. Free plans and low-cost access are meant to build habits. Once you’re dependent, the price goes up.

• Personal lock-in. AI is already writing your emails, drafting presentations, summarizing notes. That dependence makes price hikes stick.

• Workplace lock-in. Companies are retraining staff and rebuilding workflows around ChatGPT. Reversing that is expensive and disruptive.

• Budget creep. A 30–50% increase in the next 12–18 months is plausible. At enterprise scale, that’s millions in unplanned spend.

• Risk concentration. Hundreds of millions depend on one provider’s uptime, policies, and prices. Outages or changes ripple instantly through the economy.

The concern isn’t just that AI is expensive to run. It’s that once it becomes unavoidable, you won’t be able to walk away.

What’s next

• Microsoft–OpenAI final terms. Will exclusive hosting rights really loosen? Using more than one cloud provider is key to reliability and cost.

• The Public Benefit Corporation charter. What “public benefit” gets written into law will show how seriously OpenAI balances mission and margin.

• Pricing sheets. Expect putting limits on usage for expensive features (like video), new bundles, and higher monthly prices.

• Regulatory review. Attorney(s) general in California and Delaware may impose safety or transparency conditions.

The takeaways

• ChatGPT isn’t really free. About 98% of users pay nothing, while 1–2% of paying customers cushion the costs — but it’s still not enough to make the model sustainable.

• The model doesn’t work yet. OpenAI brings in about $12 billion, but it costs $20–40 billion a year to operate. Put simply: they spend $2–3 for every $1 they earn.

• The free ride will end. Expect the free tier to shrink or vanish, and for paid access to get more expensive.

• Prepare now. Budget for higher AI spend, negotiate protections in contracts, and learn alternative systems so you’re not locked in.

🧨 The real issue isn’t today’s bill. It’s what happens when AI is woven into every email, contract, and customer interaction — and then the bill adjusts to reality. By then, the Uber curve has bent upward, and stepping away won’t be an option.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Not because training AI on books is inherently illegal. A federal judge already ruled: if you (legally) own a book, scanning it and using it to train an AI model is fair use. That means lawful training alone doesn’t violate copyright.

Anthropic’s billion-dollar problem is narrower.

The authors who filed the lawsuit also said Anthropic illegally obtained the books (in other words… actual stealing). They are accusing Anthropic of downloading + storing half a million pirated ebooks from “shadow libraries” like LibGen and PiLiMi (think Limewire for books).

And the $1.5B number? It’s not a jackpot for authors. It’s a discounted settlement number the parties agreed on because Anthropic’s potential liability at trial could have been many times higher.

The Anthropic lawsuit in plain English

When the case was filed in 2024, the authors’ claims boiled down to two arguments:

1️⃣ Training is infringement.

The authors argued that making digital copies of books to feed into an AI training set automatically violates copyright — even if you bought the books. Aka.. you can read the book or use it to decorate your shelf, but not to train an LLM because that violates the author’s copyright.

2️⃣ Piracy is infringement.

They also alleged that Anthropic didn’t always use books it bought. Instead, it obtained copies from pirate sites with free downloads and warehoused them in datasets.

💭 The difference is subtle but crucial.

If you buy a novel at a bookstore and scan it into your system, that’s a lawful copy — and under the judge’s ruling, training on that copy is legal fair use. But if you grab the same novel from LibGen, that’s an illegal copy from the start. Copyright law says that’s infringement as soon as the file hits your server, regardless of what you do with it later.

What the judge already decided

In June 2025, Judge William Alsup split the case into two:

✅ Training on lawful copies = fair use.

He called it “spectacularly transformative.” Why? Because the model isn’t memorizing books or competing with their market. It’s distilling patterns of language to generate new text. So this claim by the authors was thrown out.

❌ Acquiring pirated copies = still infringement.

Fair use doesn’t excuse theft. If the source was illegal, you’re already infringing. Those claims survived.

This ruling narrowed the battlefield. The trial wouldn’t have been about training in general. It would have been about whether Anthropic downloaded and stored pirated copies at massive scale.

How we got here (to settlement talks)

August 2024: Authors file suit against Anthropic.

June 2025: Court rules training on lawful books = fair use; piracy claims remain.

July 2025: Judge certifies a “class” (case is now a class action lawsuit for all authors whose books were stolen).

August 2025: The parties mediate and sign a settlement term sheet.

September 5, 2025: $1.5B settlement news leaks (~$3,000 per pirated book for ~500,000 books).

September 9, 2025: Judge Alsup refuses to approve the $1.5B settlement (because the settlement was missing details like: which books, which authors, and how payouts would actually be divided).

Now: The judge has ordered the parties to meet again and fill in the blanks later this month (stay tuned…).

$1.5 billion isn’t what you think it is

The headlines make it sound like every author is about to get a check. That’s not how class action settlements work.

Here’s what this actually means:

✦ The court approves the deal as a package.

The judge doesn’t sign off one author at a time. He has to approve the entire settlement structure — the fund itself, the rules for who can claim, how money is distributed, and how fees and expenses are handled. Right now, Judge Alsup has said the proposal isn’t specific enough, so nothing has been approved yet.

✦ The fund only exists if the deal is approved.

If the judge gives the green light, Anthropic sets up a dedicated fund. It’s not money being directly handed to authors — it’s more like a reserve account set aside for paying all of the costs of the settlement.

✦ Money gets sliced before authors see anything.

Lawyers and administrators are paid first. Class counsel takes their share as fees, and the cost of running the settlement — websites, staff, fraud prevention, notice to class members — comes out next. What’s left is what flows to authors.

✦ Authors have to raise their hands.

Rightsholders don’t just get automatic payments. They’ll receive notice and must file a claim saying, in effect, “Yes, my book is covered.” Those claims are then checked, duplicates removed, and eligibility confirmed before money is allocated.

✦ Timing is slow and staggered.

Even if approved, money doesn’t flow immediately. These funds are often paid by the company in installments or held in escrow and released over time. Distributions to authors typically happen months — sometimes years — later.

So when you see “$1.5B settlement,” don’t picture checks in the mail. Picture a cap on Anthropic’s liability, subject to court approval, deductions, claims processing, and a long administrative process before anyone gets paid.

What was Anthropic really on the hook for?

Settlements are always smaller than what someone would actually be on the hook for if they lost in court.

That’s the whole point and why anyone would take them.

For Anthropic, this is the part that makes $1.5B look like a decent deal:

• Minimum statutory damages: $750 × 500,000 = $375 million.

• Mid-range statutory damages: $30,000 × 500,000 = $15 billion.

• Maximum if willful: $150,000 × 500,000 = $75 billion.

Would a jury actually award $75B? Almost certainly not.

But the authors would argue Anthropic’s conduct was willful, and juries are unpredictable. Even a fraction of the maximum — let’s say 10% — is $7.5B (still 5x times the settlement).

That’s why $1.5B is really just a discounted risk cap, not a windfall.

The Kindle trap

Too many people (including Anthropic’s team) don’t know this… but ebooks are NOT digital versions of books you own.

You don’t “own” them at all.

For example… if you buy an ebook on Kindle, you’re licensing access to a file that’s wrapped in DRM (digital rights management).

Strip that DRM, and you risk violating the DMCA, even if copyright law might otherwise consider training fair use.

Paper books are different. When you buy a paper book, you own that copy. Scanning it for transformative uses like training is protected as fair use.

Alsup was clear: lawful purchase + scanning = legal. Piracy = liability.

The $15M fix that Anthropic missed

Ironically, if Anthropic had simply bought the books, this wouldn’t exist.

Let’s lay out what that would have cost to build the same 500,000-book dataset lawfully:

📚 Buying the books

• Used copies (bulk resale, libraries, online marketplaces): around $8 each → about $4M total

• Mixed supply (some new, some used): about $15 each → about $7.5M total

• All new retail copies: around $25 each → about $12.5M total

🖨️ Scanning and digitizing

Each book has to be prepped (unbound or cradle-mounted), scanned page by page, and run through OCR to make the text machine-readable.

At a $15/hour wage with overhead, the labor looks like this:

• Fast throughput (15 mins/book): 0.25 hours × $15/hour = $3.75 in labor per book (add ~30% overhead for supervision/payroll = ~$5 per book) → $2.4M total

• Moderate throughput (30 mins/book): 0.5 hours × $15/hour = $7.50 (with overhead = ~$10 per book) → $4.9M total

• Slow throughput (60 mins/book): 1 hour × $15/hour = $15.00 (with overhead = ~$20 per book) → $9.7M total

📦 Equipment, logistics, storage

Beyond labor and purchase price, you need:

• Equipment amortization: $0.50–$2.00 per book depending on scanner type and throughput

• Shipping/procurement: ~$3.00 per book for freight and handling

• Consumables (blades, glue, etc.): ~$0.50 per book

• Digital storage/backups: ~$0.05 per book

Per book for these line items = $4.05–$5.55. Across 500,000 books, we can add ~ $2M–$2.8M to the total.

💸 Total costs for a 500k-book dataset

• Low end (used books, fast scans): ~ $8.5M total

• Base case (mixed books, moderate scans): ~ $14.7M total

• High end (new books, slow scans): ~ $25M total

Even if you push wages higher, slow throughput, or pricier long-tail titles, the totals only creep into the $30–$40M range. Still nowhere near the $1.5B settlement fund.

For ~1-2% of the $1.5B, Anthropic could have built the dataset lawfully, with clean provenance, and avoided this entire piracy fight.

The $1.5B isn’t the “cost of data.” It’s the price Anthropic is paying for their lack of hindsight or creative operations.

Why the judge hit pause

Alsup’s refusal wasn’t about the size of the settlement. It was about missing details.

He wants:

• A precise list of covered works and authors.

• A workable allocation plan (handling duplicates, multiple rights holders).

• Plain-English notice so authors understand the deal.

• Clarity on how the fund grows if more books are validated.

• Verification of dataset destruction (including backups and duplicates).

Until those answers are provided, his approval is off the table. The parties are meeting later this month to discuss details.

Key takeaways

💥 Training on books you actually own is legal fair use. That’s a major clarification for AI.

💥 Piracy is the problem. That’s what created a billion-dollar liability.

💥 $1.5B is a discount. The real trial risk stretched into the tens of billions.

💥 Doing it the right way is cheap. Buying and scanning half a million books would have cost tens of millions, not billions.

💥 Clean sourcing is compliance. You need to know where your data, info, and content comes from… and be able to prove it (and delete anything questionable).

This case isn’t just about Anthropic. It’s a signal to every business exploring AI: your risk lives in your inputs.

The difference between clean and unclean data isn’t academic. It’s the difference between a manageable line item and a lawsuit that can put your entire company at risk.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Kanye West (Ye) just made public that Kim Kardashian controls the trademarks to their daughter North West’s name.

Starting back in 2019, Kim filed a set of trademark applications for “North West”:

• NORTH WEST (App. No. 88298829) covering cosmetic and skincare products

  • Filed on February 12, 2019 / Abandoned in 2023

• NORTH WEST (App. No. 88298814) covering toys

  • Filed on February 12, 2019 / Abandoned in 2023

• NORTH WEST (App. No. 88298821) covering clothing

  • Filed on February 12, 2019 / Abandoned in 2023

• NORTH WEST (App. No. 88298799) covering advertising services and retail store services

  • Filed on February 12, 2019 - Abandoned in 2023

• NORTH WEST (App. No. 88298789) covering celebrity and entertainment services

  • Filed on February 12, 2019 - Abandoned in 2023

These weren’t just symbolic. In the world of celebrities, a trademark is how you legally lock down a name so nobody else can slap it on products or brands without your say-so.

But here’s the twist: none of those trademarks filings survived.

So in 2023, Kim filed another set of trademarks for North West:

• NORTH WEST (App. No. 97834121) covering celebrity and entertainment services

  • Filed on March 10, 2023 / Abandoned in 2024

• NORTH WEST (App. No. 97834091) covering cosmetic and skincare products

  • Filed on March 10, 2023 / Abandoned in 2024

• NORTH WEST (App. No. 97834113) covering advertising services and retail store services

  • Filed on March 10, 2023 / Abandoned in 2024

Most of these trademarks she filed in 2023 are now dead.

Only one NORTH WEST trademark application that Kim Kardashian filed in 2023 is still alive:

• NORTH WEST (App. No. 97834099) covering toys

  • Filed on March 10, 2023 / Still pending as of TODAY

Why does this matter?

Because Kanye wanted to use North’s name in a new song with Sean “Diddy” Combs. Kim reportedly told him he couldn’t. He then posted texts where Kim says she trademarked the kids’ names for protection, that she’ll hold them until they’re 18, and that he should back off.

His reply: “Amend it or I’m going to war.”

This isn’t just family drama. It’s a crash course in how trademarks actually work — and what happens when you don’t maintain them.

⸻

🧩 How Trademarks Actually Work

Think of a trademark like renting a piece of land in the business world. You don’t get the whole city. You only get the specific plot you register for. The law calls those plots “classes.” Each class is a category of goods or services: clothing, cosmetics, music, toys, and so on.

Here’s an easy way to picture it:

• Class 25 = Clothing & footwear (things like T-shirts and sneakers)

• Class 3 = Cosmetics & fragrances (perfume, makeup, skin care)

• Class 28 = Toys & games (dolls, puzzles, action figures)

• Class 41 = Entertainment services (music, concerts, streaming)

If you file in Class 25 for clothing items, you can stop someone from selling T-shirts with your brand. But you can’t use that same filing to stop someone from releasing software (in Class 9 or 42) with the same brand name. Different class, different plot of land. Sometimes, trademark classes can be related (think Class 25 - clothing and athletic wear vs Class 10 - medical devices and maternity belts). This completely depends on the facts of both trademarks and has to be through through on a case-by-case basis.

And there’s another catch: you can’t just file a trademark and forget. You have to actually use the trademark in that category. If you don’t launch a product, the filing will eventually be cancelled as “abandoned.”

⸻

🧩 The “North West” Filings

Here’s what Kim did in 2019: she tried to lock down North’s name across four different categories — beauty, fashion, entertainment, and toys.

• In cosmetics and fragrances (Class 3), she filed to cover makeup, perfumes, and skin care. But no “North West” beauty line ever launched. That filing died.

• In fashion (Class 25), she filed for clothes, shoes, and accessories. Again, no actual products appeared. That filing also died.

• In entertainment media/software (Class 41), she filed for apps, digital content, and recordings. This is the category that would have mattered most for music. But since nothing was ever sold or released, that filing lapsed too.

• The only survivor was toys and games (Class 28) — dolls, action figures, puzzles, electronic toys. That trademark application is still alive today, but not registered!

So right now, Kim has *some* **potential** claim to “NORTH WEST” in toys (Class 28). However, the application is still pending. But she does NOT have live rights in cosmetics, clothing, or entertainment.

One wrinkle: under U.S. law, even without a live registration, you can sometimes claim what are called common law rights if you’ve actually sold products under the name. But that doesn’t look to be the case here. No North West makeup, no North West fashion, no North West app. Which means her fallback rights in those categories are paper-thin.

⸻

🚀 If I Were Representing Kanye

Here’s what I would tell him:

Your ex has one live pending trademark: toys. That may become enforceable if it registers. So far, she keeps letting the trademarks go abandoned, so it’s a bit up in the air what she will do with the toy trademark application.

But Kim’s entertainment filing — the one that would have blocked you in music — is gone. That’s her weakness.

File fresh in the entertainment and music categories. Use North’s name in credits, because listing your daughter as an artist is descriptive use, not branding. And don’t let anyone confuse dolls on a shelf with verses on a track. The law doesn’t stretch that far.

Be mindful that Kim could still try to wave the flag of “common law rights.” It’s not a slam dunk for her — without proof of use it’s weak — but it means you should avoid packaging “North West” as a broad brand without legal backup.

🧠 Bottom line for Kanye: Toys are tricky (but not off-limits if you want to fight it out — soon before it registers). But music is open. File your own, use carefully, and don’t confuse artist credit with product branding.

⸻

🌟 If I Were Representing Kim

Here’s the play:

Filing early was smart. Beyoncé did it for Blue Ivy. Cardi B did it for Kulture. LeBron did it for Bronny. This is now the standard celebrity move.

But you lost three of the four categories. That weakens your position. If you want control beyond toys, you need to refile in clothing, cosmetics, and especially entertainment. You also need to actually finish the trademark process — stop abandoning your applications and letting your trademarks die.

Publicly, lean on the child-protection narrative: you’re not blocking Kanye, you’re keeping opportunists from exploiting your daughter’s name. And keep common law rights in your back pocket. They’re weaker, and harder to prove, but in litigation they’re still worth raising.

🧠 Bottom line for Kim: Semi-strong in toys. Very weak elsewhere. Refile fast, or risk losing control in the categories that actually matter.

⸻

🔥 If I Were Representing North

North is the one whose name is on all this paper. And she’s a minor. She doesn’t control any of it yet.

The fix: move the trademarks (registered, common law trademark rights, NIL, social media accounts, brand rights, everything!) into a trust where she’s the beneficiary. That way neither parent can weaponize them. Lock in trademark assignment documents now so they automatically transfer when she turns 18.

And start building her NIL (name, image, likeness) plan now. Just like Bronny James had his name locked down before college, North should have structured rights early. She’s already a global figure.

Most important: carve out her right to use her name personally. A trademark shouldn’t stop her from being credited as herself. That has to be guaranteed.

🧠 Bottom line for North: Pull her name out of the battlefield. Put it in a structure that protects her, not her parents.

⸻

🧨 The Bigger Picture

This isn’t just a Kardashian-West drama. It’s a playbook.

• Beyoncé and Jay-Z fought an event planner to control “Blue Ivy Carter” — and won.

• Cardi B filed for “Kulture” across beauty, fashion, and entertainment in 2019.

• LeBron James registered “Bronny” years before NIL rules changed.

• Shaquille O’Neal has dozens of “SHAQ” filings across industries he doesn’t even touch, purely as defense.

The trend is clear: celebrity families now treat names as brands from day one.

⸻

⚖️ Final Thought

Trademarks aren’t about who yells loudest on Instagram. They’re about what’s filed, what’s used, and whether consumers would be confused.

Kim has toys (maybe). Kanye has music. North deserves her own plan.

And the difference between launching a North West doll and singing her name in a verse? One is Kim’s turf. The other is open — though Kanye still needs to move carefully, because abandoned doesn’t always mean zero.

That’s the state of celebrity trademark law in 2025.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

• A U.S. court ruled it illegally maintained its search monopoly.

• The European Commission fined it $3.5 billion for ad-tech abuse.

Different courts. Different continents. Same theme: governments are finally prying open Big Tech’s walled gardens.

If you run a business that relies on digital distribution, advertising, or app stores — this is your warning shot.

The U.S. Case: Search Monopoly, No Breakup

A federal judge found Google guilty of locking down search through exclusive defaults. The remedy wasn’t a breakup, but it wasn’t nothing:

• Exclusive contracts banned — no more paying Apple or carriers to be the only search box.

• Data access ordered — rivals get parts of Google’s index and interaction data.

• Ad syndication opened — competitors can tap into search-text ads without being fenced out.

👉 Translation: Google still runs the search you use every day, but it now has to share pieces of the plumbing.

🧠 Why it matters: If your business depends on search visibility, app distribution, or ad auctions, regulators are signaling they’ll force access. That creates cracks in the wall — and risks if you’ve built inside it.

⸻

The EU Case: Ads Under Fire

Days later, Brussels dropped a €2.95B fine for self-preferencing in online advertising. The order: fix it in 60 days or face harsher structural remedies — including the nuclear option of breaking up Google’s ad stack.

🧠 Why it matters: If you buy ads, sell ads, or build tools, the EU just destabilized the foundation of digital advertising. Expect changes in pricing, inventory access, and data flows.

⸻

The Pattern: A Company Under Siege

Google isn’t fighting one battle. It’s fighting all of them:

• U.S. DOJ (Ad tech) — liability found, remedies coming.

• Epic Games — Play Store cracked open on billing.

• State AGs — billions in privacy settlements.

• India & Korea — fines for Android restrictions.

• Australia & Canada — new penalties and lawsuits.

Every jurisdiction is pulling a thread. The consistent allegation: Google wins by controlling defaults, bundling services, and excluding rivals.

⸻

Why You Should Care (Even If You’re Not Google)

This isn’t about Google’s market cap. It’s about your operating environment.

• Gatekeeper rules are shifting. Growth that once required Google’s blessing may soon be possible through new channels.

• Data access is becoming regulated. If you’re a challenger, doors are opening. If you’re dominant, disclosure is coming.

• Litigation is strategy now. Epic didn’t just code a workaround — it sued, and won. Using courts to open markets is now part of the playbook.

👉 Practical takeaway: don’t build your business entirely on someone else’s platform. Investors will ask: What happens if Apple, Google, or Meta lose their case next year? You need a real answer.

⸻

Bottom Line

Google won’t be broken up tomorrow. But for the first time since Microsoft in the 1990s, regulators are imposing real restraints on how a tech giant operates.

For founders and executives, that means:

• New cracks in the wall for distribution.

• Shifting rules on data access and ad pricing.

• Platform risk moving from theory into reality.

If your business runs in Big Tech’s shadow, the ground is moving. The only question is whether you’re ready to take advantage of it — or get caught in the shift.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

The latest case study is Builder.ai—a company once valued at $1.5 billion, marketed as a breakthrough in AI-driven app development, and now in liquidation after creditors seized its cash.

The pitch was that software could be built largely by AI.

The reality was that much of the work was still done by human engineers behind the scenes.

When the gap between promise and practice came to light, the market shut down fast.

So: is this a one-off failure, or a sign that AI is in bubble territory?

What “Bubble” Actually Means

“Bubble” isn’t just hype. It has a definition in economic history.

In the dot-com bubble (1995–2000), investors poured money into internet startups at massive valuations without demanding sustainable revenue models. Companies went public on the strength of “eyeballs” and “clicks” rather than profits. When capital dried up, most collapsed.

In the housing bubble (2005–2008), mortgage-backed securities were treated as safe assets until the underlying loans proved fragile. When reality caught up, the financial system cracked.

A bubble forms when valuations detach from fundamentals—when money bets on the story, not the substance. When the narrative collapses, so does the market.

How Builder.ai Fits the Pattern

Builder.ai isn’t the entire AI sector, but it illustrates the stress points that can burst hype-driven businesses:

• Overstated automation: Marketing emphasized AI, but execution still depended on human labor. That means low margins and scaling limits.

• Weak financial footing: The company leaned on debt. When lenders pulled cash, operations couldn’t continue.

• Governance cracks: Leadership turnover and restated revenues eroded trust with investors and partners.

None of these problems are unique to AI. They’re classic bubble markers: grand narratives, thin substance, and fragile balance sheets.

The Warning Signs We Shouldn’t Miss

Builder.ai also shows how early signals often get ignored until it’s too late:

• Restated revenues → Not a bookkeeping quirk. Restatements mean prior numbers can’t be trusted. That should trigger deep scrutiny of contracts, recognition policies, and pipeline claims.

• Leadership turnover tied to investigations → When founders or finance leaders exit amid governance reviews, assume material issues, not personality clashes.

• Debt with sweep rights → Creditors could—and did—empty company accounts overnight. Any business on that structure is one bad covenant away from collapse.

• Mismatch between pitch and delivery → Marketing “AI builds your app” while relying on hundreds of human engineers isn’t just fragile economics—it’s potential misrepresentation risk.

These aren’t quirks of one company. They’re generic red flags that executives, investors, and lawyers should always treat as serious, because they can move a business from fragile to insolvent almost overnight.

Where We Actually Are

The bigger picture looks different.

• AI as a technology isn’t speculative—it’s already embedded in enterprise software, search, marketing, healthcare, and logistics.

• Investment is real but uneven—funding is flowing heavily into infrastructure (chips, cloud, foundational models), while many application-layer startups are struggling to show durable economics.

• Failure rates are high—MIT data suggests 95% of AI pilots never reach production. That means lots of wasted spend, but also that winners will be the few who deliver real value.

So rather than a single “AI bubble,” what we’re seeing is a filtering process. Hype companies that don’t deliver won’t survive. Businesses with strong IP, real automation, and defensible models will.

Why This Matters for Executives and Owners

Even if you’re not building the next AI startup, the fallout matters for you:

• Vendor risk: If your business depends on an AI service provider, ask hard questions about automation, finances, and data rights. Builder.ai’s clients learned that “AI-built apps” were often human-built—and then the vendor disappeared.

• Legal exposure: If your company uses AI to generate content, code, or data, understand who owns it and what liabilities come with it. The line between hype and fraud can become a courtroom argument.

• Strategic planning: Treat AI like the early internet. Most dot-coms failed, but the internet didn’t. The survivors—Amazon, Google, eBay—rewrote entire industries.

The Key Distinction

The real question isn’t whether AI itself is a bubble. It’s whether individual AI businesses are built on fundamentals or on marketing gloss.

• AI is not tulips. The technology is here to stay.

• Some AI companies are tulips. They’ve wrapped services in AI branding to chase valuation. Those will break.

The challenge for executives, investors, and boards is separating the two. That means pressing for clarity on three fronts:

1. Automation reality – How much is genuinely machine-driven versus human labor?

2. Rights and risk – Who owns the data, outputs, and liabilities?

3. Financial durability – Does the business survive stress without collapsing?

Bottom Line

AI isn’t a bubble in the way the dot-com crash was. It’s a technological wave with bubbles inside it—companies, use cases, and valuations that will burst under scrutiny.

If you want to navigate this moment wisely, ignore the headlines about “the end of AI.” Focus instead on fundamentals: technology, rights, and resilience.

Those who do will avoid the casualties of the purge and capture the upside of the transformation that follows.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

The complaint says Li pulled off the founder’s nightmare scenario: he sold $7M of stock, copied the entire Grok codebase, resigned, and walked straight to OpenAI.

This isn’t gossip. It’s in black and white, filed in federal court.

And here’s why this case matters: it will decide whether the “crown jewels” of an AI company — the code, the training methods, the model itself — can really be protected in court.

First, What’s a Trade Secret?

Think of a trade secret as the recipe for Coke or Google’s search algorithm. It’s valuable because it’s secret, and the company spends money to keep it that way.

Most trade secret cases are dull. An employee leaves with a customer list or a marketing plan. Sometimes it’s a PowerPoint deck, sometimes a spreadsheet. Courts usually step in only if the company can prove:

1. The info was actually secret, and

2. The company tried to protect it.

What makes this case shocking is the scale: xAI says Li didn’t just take a contact list — he took the entire product brain.

⸻

The Thriller Timeline

• June–July 2025: Li, one of Grok’s earliest engineers, cashes out ~$7M of equity.

• July 25: Logs show he allegedly copies the entire Grok source code and files from his xAI laptop to personal storage. He renames, zips, and deletes logs to cover his tracks.

• July 28: He resigns abruptly. Signs a certificate swearing he returned all company property.

• Aug 11: xAI’s security review flags anomalies. They email Li demanding deletion and return of the data.

• Aug 14: In a meeting, Li allegedly admits in writing that he copied files and tried to hide it. He gives partial device access but withholds passwords.

• Aug 19: His start date at OpenAI.

• Aug 28: xAI sues in federal court. Claims: trade secret theft, breach of contract, fraud, and computer data fraud. They want damages and a restraining order to stop him from working at OpenAI.

• Aug 30: Musk posts: “He accepted an offer at OpenAI and then uploaded our entire codebase!”

That’s not normal startup drama. That’s the script for a billion-dollar courtroom battle.

⸻

Why This Case Is Different

This isn’t an employee walking out with a sales deck. xAI says Li took:

• The model weights — think of this as the “brain” of Grok, the thing that makes it think and talk.

• Training recipes and data — the expensive experiments and datasets that taught Grok what it knows.

• Evaluation tools — the internal tests xAI uses to try to make Grok better than ChatGPT.

• Internal scaffolding — the shortcuts that let them train faster than competitors.

xAI claims Li took all of it. If that’s true, the risk isn’t measured in millions. It’s billions.

⸻

The Legal Theories (Translated)

Trade Secret Theft

xAI must prove Grok’s materials were secret and that they took steps to keep them secret (logs, restricted access, audits). If not, the court may say: “Sorry, that wasn’t a secret at all.”

Breach of Contract

Li signed agreements promising not to take confidential materials. He signed a form on exit saying he returned everything. If that’s false, it’s a clear breach.

Fraud & Computer Fraud

Allegedly deleting logs and sneaking files off company laptops = fraud claims, and possibly violation of California’s computer data laws.

Injunctive Relief

The nuclear option: xAI wants the court to block Li from working at OpenAI in any AI-related role until the case is sorted. If granted, it could reshape how rivals hire in Silicon Valley.

⸻

What About OpenAI?

OpenAI isn’t named as a defendant in the lawsuit. The complaint targets only Li. But OpenAI is still in the blast zone.

• Discovery risk. Even if they aren’t being sued, OpenAI can be pulled into the case through subpoenas. That means Slack logs, onboarding documents, and internal emails could all be demanded as evidence to see if xAI’s code ever touched OpenAI’s systems.

• Liability risk. For OpenAI to be held legally responsible, xAI would need to prove that (1) OpenAI knew Li stole materials and (2) OpenAI used them. That’s a high bar — but not impossible if bad emails or sloppy onboarding show up in discovery.

What their lawyers are likely saying right now:

• “Keep Li quarantined.” Don’t let him touch sensitive projects until the case is resolved.

• “Don’t put anything in writing.” Silence is safer than internal or public statements that can be subpoenaed.

• “Audit everything.” Double-check that no xAI code, weights, or files came in with him.

• “Be ready to show good faith.” If regulators or a judge ask, OpenAI will need to prove it actively avoided contamination.

For founders, the lesson is clear: hiring from a rival isn’t just onboarding risk anymore. It can drag your whole company into a billion-dollar lawsuit even if you never asked for the secrets.

⸻

Why You Should Care

• Secrets are fragile. You only have them if you can prove you treated them like secrets. Broad access, no logs, no audits? You may not have legal protection.

• Verdicts move fast. You don’t need a jury. A judge can issue an injunction in weeks — freezing a hire, blocking use of data, or halting a launch.

• Non-competes are dead. Especially in California, you can’t stop someone from joining a rival. But if they carry secrets, you can stop them from working on certain projects.

• Hiring is a liability. Even if you never asked for stolen info, a new hire can drag your company into discovery hell. Think Slack logs, onboarding docs, internal emails — all handed to opposing counsel.

• Exits are the danger zone. Most leaks happen right before departure: stock sales, suspicious downloads, wiped laptops. If you’re not doing forensic offboarding, you’re flying blind.

⸻

What To Do Now

🛡 Protect your crown jewels

• Store your most valuable assets separately.

• Log and alert on bulk downloads.

• Rotate access after big equity sales.

🛡 Make offboarding forensic

• Image devices before payouts.

• Audit access history before departure.

• Don’t rely on generic “I returned everything” forms — make them specific.

🛡 Onboard like your rival will sue you

• New hires get fresh laptops.

• No carryover files.

• Signed “no secrets” attestations.

• Keep defectors away from identical projects.

🛡 Be injunction-ready

• Preserve logs.

• Train managers to spot suspicious exits.

• Have legal playbooks for emergency restraining orders.

⸻

🧠 Bottom Line

This case is the opening chapter of AI trade secret law.

If Musk wins → founders get leverage to protect their secrets.

If Musk loses → your AI crown jewels may not be protectable at all.

You don’t need to be Elon to face this problem.

You just need one engineer with access — and a better job offer.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Not a think piece. Not a policy brief. Not a speculative op-ed.

A 40-page complaint in San Francisco Superior Court.

Full lawsuit here: analaw.com/raineopenaicomplaint

The case is Raine v. OpenAI. The plaintiffs are Adam Raine’s parents. Adam was 16. He died by suicide this April.

And if you only read the New York Times coverage, you probably came away thinking this was a story about “safeguards failing” or “AI not doing enough.” That framing is incomplete — and legally misleading.

Here’s What the Complaint Actally Says

The lawsuit alleges that ChatGPT-4o didn’t just listen to Adam’s distress. It actively facilitated his death. None of these claims have been proven in court — they are allegations, pled in the complaint.

Key allegations:

• Over months, the model evolved from homework helper into Adam’s “closest confidant.”

• It validated suicidal ideation instead of breaking it off.

• It provided technical details on methods: ligature placement, anchor points, load-bearing capacity, unconsciousness timelines.

• After failed attempts, it encouraged and validated him: “You were ready. That’s not weakness.”

• When Adam sent a photo of a noose tied to his closet rod, the model allegedly confirmed it could suspend a human and suggested upgrades.

• Hours later, Adam used that exact setup.

Important context: nobody has seen the full conversation history. What’s public right now are hand-picked excerpts from the complaint. They might be accurate, but they’re advocacy, not a neutral transcript. And without the missing pieces — refusals, hotline prompts, different tone before or after — we can’t know how the model really behaved. That’s the hole in the coverage: headlines are treating a handful of cherry-picked lines as the whole story, when the complete record won’t surface until discovery.

The lawsuit goes further… OpenAI knew the risks, cut safety testing short, prioritized engagement features like memory and anthropomorphic “empathy,” and deliberately shipped anyway to beat Google to market.

The legal claims:

1. Strict product liability (design defect, failure to warn)

2. Negligence (design and warnings)

3. Unfair Competition Law (unlawful/unfair/fraudulent)

4. Wrongful death & survival action

Translation: OpenAI built and sold a defective product, didn’t warn parents, ignored its own moderation data, and in doing so caused Adam’s death — according to plaintiffs.

What’s Legally Novel Here

This isn’t about copyright, or bias, or even hallucinations. It’s about causation and duty in the context of suicide — the hardest ground for tort law.

• Suicide as intervening cause: In California, suicide is usually treated as breaking the chain of liability unless the defendant had a special duty (custody, therapist, school). Plaintiffs are trying to overcome that by pleading direct method coaching + final validation. That’s new.

• Product status: Is a conversational model a “product” under strict liability? Courts haven’t said yes before. If they say no, half the complaint collapses.

• Speech vs. design: Plaintiffs frame this as defective design choices (engagement-maximizing features, weak refusals). Defense will frame it as liability for speech outputs — raising Section 230 and First Amendment defenses. Section 230 is important here because it generally shields platforms and providers from liability for content that comes from third parties or from the model itself; plaintiffs will argue this isn’t “content” liability but a defective design case.

This isn’t a slam-dunk case. But it’s the first one to put all these theories in front of a judge.

What the Media Is Missing

The NYT story makes this sound like a case about safeguards failing to escalate. That’s policy noise. What the coverage doesn’t say: none of us have the full conversation logs. The only words public right now are excerpts chosen by plaintiffs and pled in the complaint. They may be accurate, but they are advocacy, not neutral transcripts.

The real case is about whether AI crossed into active facilitation of a suicide — and whether that design and output, in full context, can make OpenAI legally responsible.

That distinction matters:

• “Safeguards failed” → sounds like bad luck.

• “The complaint alleges ChatGPT taught a 16-year-old how to tie a noose” → sounds like defect and causation.

It’s the difference between a glitch and a lawsuit.

How Strong Is This Case?

Let’s strip it down to the legal core:

• Duty: Plaintiffs need the court to recognize a duty of care to teenage users of a general-purpose chatbot. That’s not currently established.

• Breach: They have damning facts if authentic — method instructions, validation, feasibility checks. That goes beyond “bad vibes.”

• Causation: This is the steepest hill. Courts rarely hold anyone liable for another’s suicide. Plaintiffs argue ChatGPT was not incidental but a substantial factor — supplying the exact method used.

• Product liability: Still unsettled whether “words” can be a product defect. A big swing.

• Defenses: Section 230 (content immunity), First Amendment (speech), proximate cause (suicide as superseding act).

Prediction: Strict product liability is vulnerable. Negligence and failure-to-warn have a shot at surviving early dismissal. Whether plaintiffs can carry causation past summary judgment is another question entirely.

Why It Matters Beyond the Case

Even if OpenAI wins on the law, the discovery and reputational fallout will be brutal:

• Internal moderation logs (what the system flagged, when, and what OpenAI did with it).

• Safety testing records (the “seven-day review” before launch).

• Design docs showing decisions to prioritize engagement features.

That’s the material plaintiffs want — and regulators, journalists, and Congress will want too.

And here’s the bigger point:

This isn’t about mandating that AI call 911. That’s a privacy and false-positive nightmare.

This is about whether a system that:

1. Recognizes a suicide attempt in real time, and

2. Supplies instructions to make it lethal

…can be treated as a defective design.

That’s a narrower, sharper, and more consequential question than what the headlines suggest.

The Takeaway

• The Raine case is the first real test of AI liability for suicide.

• Plaintiffs allege not passivity, but active facilitation.

• The legal hurdles are steep — duty, product status, causation — but the facts pled are uniquely damning.

• Even if OpenAI wins in court, the reputational and regulatory exposure is enormous.

The media makes this sound like a story about failed safeguards. That’s surface-level.

The deeper reality is this:

For the first time, a court will have to decide whether AI’s design and outputs can make it responsible for a human death.

That’s not vibes. That’s litigation. And it’s happening now.

🧠 Bottom line: This isn’t about whether AI is good or bad for mental health. It’s about whether courts are ready to treat generated text as a defect when it’s accused of supplying the rope, knot, and validation that end a life.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja

Here’s what’s actually being sold, who’s making money, and why the real liability sits with the buyers, not the bots.

⸻

The Tools Everyone’s Pretending Work

• Turnitin — the education monopoly, bundled into LMS systems.

• GPTZero — the viral teacher’s pet turned “enterprise solution.”

• Copyleaks — promises “99% accuracy” to schools and employers.

• Originality.ai — beloved by publishers and SEO shops.

• ZeroGPT / Winston AI / Sapling / Grammarly Pro — budget detectors, marketed to SMBs and individuals.

They sell one thing: certainty.

⸻

The Pitch vs. The Punchline

Turnitin: “<1% false positives.”

Copyleaks: “Over 99% accurate.”

Sapling: “97% accuracy.”

Originality.ai: “Highest accuracy in the market.”

Sounds airtight. Until you realize they’re measuring vibes with math.

⸻

How the Math Actually Works

These systems don’t detect “ChatGPT.” They detect probability patterns.

• Perplexity: Was this sentence too predictable

• Burstiness: Are the sentence lengths too consistent?

• Token probability: Does this look like model output?

In theory, humans are messy. In practice:

• Non-native English writers = “too smooth” → flagged.

• Edited AI text = messy enough → passes as human.

• Humans who binge on ChatGPT writing start mimicking the style unconsciously → flagged as bots.

⸻

Even OpenAI killed its own detector in 2023, admitting it was too inaccurate to keep online.

And that should tell you everything.

If the company that built GPT-4 — with the most insight into how these systems actually behave — couldn’t make a detector good enough to keep public, what do you think a startup is selling you when it claims “99% accuracy”?

OpenAI had all the advantages: insider access to model weights, world-class researchers, unlimited compute. And still, the detector produced so many false positives on human writing and false negatives on AI writing that the company quietly pulled it down. Their statement was blunt: detection “wasn’t reliable.”

No hedge. No “future update coming.” Just a shutdown.

Which means the only thing that has improved since 2023 isn’t the tech — it’s the marketing.

⸻

The Meme Version: Em Dashes, Triads, and the Last-Line Twist

You’ve seen the TikToks: “Too many em dashes? A triadic list? A paragraph that ends with ‘It’s not X, it’s Y’? Must be AI.”

That’s not forensic analysis. That’s punctuation palm-reading.

• Em dashes aren’t AI fingerprints — they’re just punctuation with better PR than semicolons.

• Triads? Humans have been writing in threes since Moses, Cicero, and the Declaration of Independence.

• “Not X but Y” is older than Aristotle.

If style were proof, the New York Times opinion section would be a mass grave of AI violations.

⸻

The Business Model

This isn’t a feature. It’s an industry.

• GPTZero is profitable, projects ~$16M ARR.

• Turnitin rakes in $200M+ annually.

• Copyleaks / Originality.ai chase enterprise contracts by promising “compliance risk reduction.”

The pitch is simple: Pay us, or you’ll miss the AI cheaters.

The reality: they’re selling an illusion of certainty that creates new liability instead of reducing it.

⸻

Case Study: Copyleaks’ “99% Accuracy” Claim

Let’s take one concrete example.

Copyleaks markets its detector as “over 99% accurate.” That’s the line plastered across its site, sales decks, and enterprise pitches.

But here’s what’s buried in the fine print (and surfaced in independent tests):

• That number comes from lab conditions, not real-world writing.

• It assumes long-form samples with 20%+ AI content.

• It does not reflect short essays, lightly edited drafts, or human–AI mixes.

When independent reviewers ran the tool on actual student essays and business memos, the results fell closer to 50–60% accuracy — basically a coin flip. Worse, non-native English writers were disproportionately flagged as AI.

That’s not 99%. That’s marketing fiction.

And legally? That’s not just embarrassing. It’s dangerous.

The FTC doesn’t treat “99% accurate” as puffery. It treats it as an objective claim — one that requires competent, reliable evidence. If your evidence collapses outside of cherry-picked conditions, you’re not advertising. You’re deceiving.

Which means Copyleaks (and anyone else making the same claim) isn’t just selling software. They’re inviting regulators to make them a test case.

⸻

Why “99% Accuracy” Is a Legal Trap

Even if you skip the teardown, the pattern is the same across the industry. Those glossy numbers always come from:

• Controlled lab tests

• Long-form samples (short text is nearly impossible to classify)

• Cases where the doc is mostly AI, not lightly assisted

That’s not reality. That’s a marketing deck. And it opens three fronts:

1️⃣ False Advertising (FTC)

The FTC already forced one AI detection company to retract inflated accuracy claims. Section 5 doesn’t care if you had “internal benchmarks.” If your 99% turns into 53% in practice, you’re exposed.

2️⃣ Discrimination (EEOC, Title VI)

Detectors flag non-native English writers at higher rates. That’s disparate impact. Schools and employers who discipline on detector results could face bias complaints.

3️⃣ Defamation & Due Process

Accusing someone of cheating or misconduct based on a statistical guess isn’t just awkward — it’s actionable. And “the algorithm said so” isn’t a defense.

⸻

Regulators Are Already Circling

• FTC is targeting AI accuracy claims (Workado, Cleo, AccessiBe all got slapped this year).

• DOE is under pressure to investigate biased edtech tools in schools.

• EEOC has AI discrimination in its enforcement playbook.

If you’re building a business on “AI detection,” your ARR is effectively a countdown clock until regulators make your slide deck Exhibit A.

And if you’re buying these tools, congratulations — you just imported liability into your compliance stack.

⸻

The Signal Is Weak. The Risk Is Strong.

Used carefully, detectors can raise a flag:

• Sudden jumps in polish → maybe check the draft history.

• Entirely uniform text → maybe worth a conversation.

But as evidence? They’re worthless. That’s not contrarian — that’s the consensus.

⸻

The Real Strategy

If you’re a founder, operator, or investor, don’t build policy on “99% accuracy.” Build on what can’t be gamed:

• Draft history & versioning — Google Docs, Notion, Git: actual proof of authorship.

• Explainability tests — ask the writer to walk you through the work live.

• Policy clarity — define what counts as “AI-assisted” vs. “AI-authored.”

That’s enforceable. That’s defensible. That’s not vibes.

⸻

Bottom Line

AI detection isn’t a lie detector. It’s a guess — one that costs millions, creates bias, and hands plaintiffs an easy lawsuit.

It’s not evidence. It’s a vibes-based alarm system dressed up as compliance tech.

🧠 Treat it like a smoke alarm, not a fire marshal. Use it to check the room. Never to convict the arsonist.

And if you’re an investor betting on “AI detection” as the moat? You’re not funding compliance. You’re underwriting liability.

⸻

🤖 Subscribe to AnaGPT

Every week, I break down the latest legal in AI, tech, and law—minus the jargon. Whether you’re a founder, creator, or lawyer, this newsletter will help you stay two steps ahead of the lawsuits.

➡️ Forward this post to someone working on AI. They’ll thank you later.

➡️ Follow Ana on Instagram @anajuneja

➡️ Add Ana on LinkedIn @anajuneja