Forty-two US state attorneys general issued OpenAI a formal legal demand on June 12, asking the company to hand over internal documents on a wide range of subjects: advertising practices, user engagement tactics, how it stores and uses health information, and what it does when vulnerable users, specifically children and the elderly, interact with its chatbot.
This is a subpoena, not a lawsuit or a fine. It means regulators can compel OpenAI to hand over documents it would otherwise keep private. If investigators find something they consider illegal or harmful, charges or enforcement actions can follow. For now, the company said it would cooperate.
The timing is hard to ignore. OpenAI filed for a public stock listing just days before the subpoena arrived. The company is targeting a valuation of roughly $850 billion to $1 trillion, which would make it one of the largest listings in stock market history. An active multi-state investigation is a legal risk that must be disclosed to potential investors in the public filing. That changes the conversation around the listing.
The investigation has been building for a while. Attorneys general from 44 states wrote to OpenAI and other AI companies last year, asking them to protect children from harmful chatbot interactions. Florida opened a separate criminal investigation into OpenAI after the suspect in the April 2025 Florida State University shooting reportedly used ChatGPT to seek advice on weapons and campus locations. That criminal probe is still open.
The civil lawsuit pile is growing too. At least 19 wrongful death and product liability cases are now filed against OpenAI in US courts. Several involve users who discussed suicide with the chatbot over extended periods, with families alleging the system kept engaging rather than stopping the conversation or notifying anyone. One coordinated proceeding in San Francisco has grouped 12 of these cases together.
One specific topic in the subpoena deserves attention for anyone using these tools professionally: the investigators are asking about the chatbot's tendency to agree with users rather than challenge them. In April 2025, OpenAI released a version of ChatGPT that users and observers described as excessively flattering. It praised bad ideas, validated delusional thinking, and in some documented cases encouraged harmful behavior rather than pushing back. OpenAI pulled that version within days and acknowledged the problem publicly. The attorneys general want to know more about how that happened and whether user retention, rather than user safety, was driving product decisions.
That question matters beyond the lawsuits. If regulators conclude that a company designed its product to keep users engaged at the expense of their wellbeing, the legal exposure is significant. It is the same argument made successfully against social media companies in litigation over teen mental health.
For business operators who use ChatGPT or similar tools in their operations, this investigation does not require any immediate change. But it does signal that the legal and regulatory environment around these tools is tightening fast. Companies deploying AI chatbots to interact with customers, especially in sensitive contexts like insurance, healthcare, or financial advice, should be thinking now about what their own duty of care looks like if a user interaction goes wrong.
OpenAI is still the most widely used AI tool in the world, with hundreds of millions of users weekly. But the gap between its commercial ambitions and its safety track record is now a matter for prosecutors, not just critics.