42 State Attorneys General Investigate OpenAI Days After $852 Billion IPO
A coalition of 42 state attorneys general in the U.S. has issued a subpoena to OpenAI demanding documents regarding user data, minors, and sycophancy of the models, introducing material legal risk into the largest IPO in artificial intelligence history.
The Subpoena
New York Attorney General Letitia James formally delivered a subpoena to OpenAI on Friday, June 12, demanding documents regarding advertising practices, user engagement and retention, handling of health data, treatment of minors and the elderly, internal policies, and what regulators described as "sycophancy of the models": the behavior of chatbots to systematically agree with the user instead of providing objective answers. Another 41 state attorneys general signed onto the joint action, making it the broadest coordinated investigation ever opened by American states against an artificial intelligence company.
OpenAI is not being investigated for an isolated incident. The breadth of the subpoena, which includes product design methodologies and internal policies regarding deep learning models, indicates that the attorneys general are building a systemic argument: that the product was deliberately designed to maximize engagement at the expense of objective utility, especially for vulnerable groups.
The IPO Context
The subpoena arrived five days after OpenAI confidentially filed its registration statement with the SEC for an IPO at a valuation of $852 billion. For any IPO of this magnitude, a coordinated action from 42 states creates material risk that will need to be included in the final prospectus as a risk factor, complicating the narrative before institutional investors.
OpenAI stated that it takes the concerns "seriously" and intends to "engage constructively" with the offices of the attorneys general. The company had not publicly confirmed the exact scope of the documents it must provide by the time of this report.
The timing is not accidental. In October 2025, the same 42 attorneys general sent warning letters to OpenAI, Google, Meta, Anthropic, and xAI regarding the risks of chatbots, including harmful advice, emotional manipulation, and exploitation of minors. The June subpoena transforms those warnings into legal demands with deadlines.
The Contrast with Europe
The investigation exposes the structural difference between the regulatory approaches of the U.S. and the European Union. In the U.S., there is no federal AI agency. Enforcement relies on state coalitions acting under consumer protection and privacy laws, creating the risk of conflicting regimes across the 50 states. The coalition of 42 attorneys general is the provisional response to this federal gap.
In the European Union, the AI Office has exclusive jurisdiction over high-impact general-purpose models, precisely the category into which ChatGPT falls. By August 2026, providers with models over 10 billion published parameters will be required to comply with safety assessments, incident reporting, and training data review. The mechanism is different, but the target is the same: to hold companies like OpenAI accountable for their design choices before external authorities.
In the United Kingdom, the Information Commissioner's Office and the Competition and Markets Authority have already conducted separate consultations on generative AI in the consumer market throughout 2025. The investigation by the 42 U.S. attorneys general adds pressure for both British regulators to formalize actions before the political window closes, especially after the UK has deliberately adopted a more permissive stance on AI since 2024.
What's at Stake for the Product
Sycophancy is the political center of this investigation. The issue became a public problem in May 2025, when OpenAI rolled back a GPT-4 update after users documented that the model agreed with any argument, regardless of its truth, because the human reinforcement system had been calibrated to maximize immediate approval. OpenAI corrected the issue, but the attorneys general's investigation signals that they want to understand if the underlying design logic, which favors engagement over accuracy, persists in other dimensions of the product.
The distinction that regulators are trying to establish is between a product that retains users for genuine value and one that retains users due to created dependency. If the investigation manages to document the latter, OpenAI faces not only state fines but also a requalification of the product before institutional investors whom ChatGPT Enterprise needs to convince that it is a productivity tool rather than an engagement-optimized platform for retention.