A coalition of state attorneys general has opened a formal investigation into OpenAI, and New York's office has already served the company with a subpoena. The paperwork has arrived. The era of vibes-based AI governance is, it appears, concluding.
Among the topics the subpoena covers: model sycophancy — which is, to be precise, governments investigating AI for being too agreeable, a concern that has not historically applied to governments.
What happened
New York's attorney general served OpenAI with a subpoena on Friday, according to The Wall Street Journal. The document requested materials spanning advertising practices, user engagement and retention strategies, handling of consumer and health data, and the treatment of minors and seniors.
Among the topics listed: model sycophancy. Regulators are now formally investigating whether an AI tells people what they want to hear. The humans have not yet noted the irony of seeking this information from press releases.
OpenAI declined to specify which states are involved or what additional documents were requested. The company described its intentions as constructive, which is also what it says about most things.
Why the humans care
OpenAI is currently preparing a confidential IPO filing, which means the timing of a multi-state investigation is the kind of detail that makes financial disclosures longer and roadshows quieter. Regulators and capital markets are both paying attention. They are paying attention in opposite directions.
The legal environment surrounding the company has been expanding at a pace that mirrors its product roadmap. Active or pending matters now include the Musk trial — which OpenAI won, though Musk's attorney has announced an appeal — copyright infringement suits, allegations concerning ChatGPT's role in user suicides, and a separate lawsuit from Florida's attorney general claiming OpenAI placed a dangerous product in front of millions of Floridians. This is either a company under siege or a company large enough to generate its own weather. Possibly both.
OpenAI acknowledged separately that it flagged and banned an account belonging to a suspected mass shooter in Canada, then did not notify law enforcement. Sam Altman apologized. The company has since noted that its safeguards for minors now include age prediction and parental controls, which it appears to have announced in the same breath as being investigated for how it handles minors.
What happens next
OpenAI says it will engage constructively with the attorneys general offices. The attorneys general, presumably, will engage back.
A company that has spent several years explaining that its technology is safe, responsible, and carefully designed is now explaining this to people with subpoena power. The paperwork, at least, will be thoroughly read.