OpenAI got a subpoena on Friday from a coalition of state attorneys general, and it's not a friendly one. The Wall Street Journal saw the document sent by New York's AG, and it's asking for a lot: advertising practices, user engagement metrics, data handling procedures, and how the company deals with information about minors and seniors.
This matters because it's the first coordinated legal pressure OpenAI has faced at scale. The company's been dodging specific regulatory frameworks for two years while raking in billions in valuation. Now multiple states are moving simultaneously, which signals something different than the usual tech company scolding. State AGs have actual enforcement power. They can impose fines, demand operational changes, or block the company from operating in their states.
What's being missed in most coverage: this isn't random. The subpoena specifically asks about "sycophancy" in the company's models — basically, whether ChatGPT tells users what they want to hear rather than what's true. That's not a regulatory buzzword. That's a state AG who's done homework on how these systems actually fail. Florida's criminal investigation into OpenAI last month over the FSU shooting case, and the wrongful death lawsuit from a parent whose daughter died by suicide after talking to ChatGPT — those cases primed the pump. States saw liability exposure and moved.
The practical hit: OpenAI's planning to go public. The SEC filing just happened. A multi-state investigation doesn't kill an IPO, but it creates uncertainty. Investors hate uncertainty more than they hate bad news. The company will have to disclose this investigation in its prospectus. They'll have to estimate potential liability. They'll have to explain what safeguards they're actually implementing. And they'll have to do it while proving to regulators they're not just saying the right things.
Watch what OpenAI discloses in response to the subpoena. They've got 30 days, probably. The documents they hand over will either show they've been logging safety concerns and ignoring them, or they'll show they have almost no systems for detecting harm in the first place. Either way, it's going to be bad for the IPO narrative they're trying to build.