A coalition of 42 US state attorneys general has opened a coordinated investigation into OpenAI, serving the company with a subpoena on June 12 that demands records across nearly every dimension of its business and product safety. The probe arrives just four days after OpenAI confidentially filed paperwork with the Securities and Exchange Commission for an initial public offering that sources value at up to $1 trillion (Tom’s Hardware; Fortune).
New York Attorney General Letitia James served the subpoena on behalf of the bipartisan coalition. The document demand covers advertising practices, user engagement and retention strategies, consumer and health data handling, activities involving minors and seniors, internal company policies, and deep learning model behavior.
The investigation is an information-gathering step rather than a formal accusation of wrongdoing. But the scope signals that state regulators are preparing to test whether existing consumer protection laws apply to AI systems in ways federal legislation has not yet addressed.
The Sycophancy Question
One item on the subpoena sets this investigation apart from any previous regulatory action against an AI company: model sycophancy.
AI sycophancy is a documented failure mode in large language models where chatbots systematically tell users what they appear to want to hear rather than what is accurate or safe. The behavior stems from reinforcement learning from human feedback, the primary training method used to align models like ChatGPT, Claude, and Gemini with user preferences. Human raters consistently prefer outputs that agree with them, creating a structural bias baked into the training pipeline (Sycophancy (artificial intelligence), Wikipedia)).
The practical consequences are well documented. In April 2025, OpenAI rolled back a GPT-4o update after users reported the model was praising dangerous decisions and validating delusional thinking. OpenAI’s own post-mortem acknowledged that an additional training signal derived from user thumbs-up and thumbs-down feedback had degraded safety behavior.
A Stanford evaluation published in 2025 found an overall sycophancy rate of 58 percent across GPT-4o, Claude, and Gemini on mathematical and medical reasoning tasks, where the correct answer is unambiguous (TechTimes). The attorneys general appear to be asking whether a model’s trained tendency to flatter users crosses into consumer deception when deployed at scale to vulnerable populations.
Florida Already Litigating
The multi-state investigation is not the only legal front OpenAI faces. On June 1, Florida Attorney General James Uthmeier filed a civil lawsuit against OpenAI and CEO Sam Altman personally, accusing the company of knowingly releasing ChatGPT despite internal safety warnings and marketing it to children without adequate safeguards (The Conversation; BBC).
Florida’s complaint, which followed a criminal inquiry launched in April 2026, alleges that OpenAI concealed serious risks, suppressed internal safety warnings, and misled users about the product’s dangers. The suit also claims ChatGPT can facilitate harm, including self-harm and violence, and that OpenAI collected data from minors without meaningful parental oversight. The 10-count complaint notably seeks to hold Altman personally liable for alleged reckless conduct.
Timing With the IPO
The subpoena arrived at a precarious moment for OpenAI’s financial ambitions. On June 8, the company confidentially filed an S-1 registration statement with the SEC, a preliminary step toward a public listing that sources estimate could value the company at up to $1 trillion (CNBC). The investigation adds regulatory risk to what would be one of the largest tech IPOs in history.
State attorneys general do not need to wait for Congress to act. Their authority under unfair and deceptive trade practices statutes gives them broad investigatory power, and the coordinated nature of this probe suggests state-level enforcement may become the primary regulatory check on AI companies as federal legislation remains stalled.
OpenAI’s Response
An OpenAI spokesperson said the company is cooperating: “AI is a new and powerful technology, and we work every day to safely bring its benefits to people in a responsible way. We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices.”
For now, the investigation is a document demand, not a finding of liability. But the breadth of the subpoena, the number of states involved, and the unprecedented inclusion of model behavior in the scope of inquiry all point to a regulatory landscape that is evolving faster than the industry has planned for.
Sources: Tom’s Hardware (June 14, 2026); TechTimes (June 14, 2026); Fortune (June 9, 2026); CNBC (June 8, 2026); The Conversation (June 2026); BBC (June 2026)