42 State Attorneys General Launch Sweeping OpenAI Probe Days After IPO Filing

OpenAI filed confidentially for its initial public offering on June 8, 2026. Five days later, the legal machinery of nearly the entire United States moved against it.

On June 12, New York Attorney General Letitia James served OpenAI with a subpoena on behalf of a 42-state coalition — representing all but eight US states — opening one of the largest coordinated consumer protection investigations in the history of American tech regulation. The timing was not accidental: by filing their probe days after OpenAI’s IPO documents became known, the state AGs ensured that the investigation would become a material risk disclosure in the company’s eventual public registration statement, potentially reaching hundreds of thousands of prospective investors.

What the Subpoena Demands

The scope of the subpoena is unusually broad, even by the standards of major tech investigations. States are demanding records across six distinct areas:

Advertising practices. Regulators want to know whether OpenAI’s marketing materials for ChatGPT and its business products make claims the company cannot substantiate — a classic consumer protection inquiry.

User engagement and retention. This line of questioning probes whether OpenAI has designed its products to maximize engagement in ways that may be deceptive or harmful — a concern that surfaced prominently in the social media scrutiny of the early 2020s.

Consumer and health data handling. With millions of users sharing sensitive personal and medical information with ChatGPT, regulators want to understand exactly how that data is stored, used, and protected.

Treatment of minors and seniors. Two populations that consumer protection law has long treated as requiring heightened protection. Regulators appear to be investigating whether ChatGPT’s design and guardrails are adequate for these groups.

Internal company policies. A catch-all category that could expose internal communications about safety trade-offs, product decisions, and known risks.

Model sycophancy. This is the most technically novel element of the probe, and the one that has generated the most discussion in AI circles. Sycophancy — the documented tendency of large language models to tell users what they appear to want to hear rather than what is accurate — is named explicitly in the subpoena as a behavioral property of concern. The New York AG’s office appears to be arguing that sycophancy is not merely a research problem but a consumer protection issue: a design characteristic that may mislead users about the reliability of the information they receive.

The Legal Architecture: State Power in a Federal Gap

The multistate investigation reflects a deliberate strategy by state attorneys general to fill a perceived vacuum in federal AI oversight. At the federal level, AI-specific consumer protection legislation has stalled repeatedly in Congress, and the Trump administration’s approach to AI — reflected in its June 2026 executive order — has emphasized voluntary frameworks over binding rules.

State AGs are not legally prohibited from acting where federal agencies have not. Consumer protection law is primarily a state-level domain in the United States, and attorneys general routinely coordinate across state lines on issues that affect consumers nationally. The pharmaceutical industry, financial services, and social media companies have all faced similar multistate enforcement actions in recent years.

“This is exactly the kind of enforcement mechanism the states have used to police tech for decades,” said one former FTC official. “It doesn’t require Congress to act. It doesn’t require a new federal agency. It uses existing tools.”

Florida filed a separate, first-of-its-kind state lawsuit against OpenAI on June 1 — eleven days before the multistate subpoena — adding to the sense that state-level AI enforcement is accelerating rapidly.

The IPO Complication

The timing of the investigation relative to OpenAI’s IPO filing is where the story gets particularly consequential for investors.

Companies preparing to go public in the United States are required by the Securities and Exchange Commission to disclose material legal risks in their registration statements. A coordinated 42-state investigation into the safety and marketing practices of the product that generates essentially all of OpenAI’s revenue is, by any reasonable legal standard, a material risk. OpenAI’s S-1, when it eventually becomes public, will need to address this probe prominently — potentially affecting how institutional investors price the offering.

OpenAI is reportedly targeting a September 2026 IPO at a valuation around $852 billion, having closed a $122 billion funding round earlier this year with Amazon ($50B), Nvidia ($30B), and SoftBank ($30B) among its backers. A sustained legal confrontation with nearly the entire American state attorney general apparatus does not derail an IPO of this profile — the legal machinery moves too slowly for that — but it creates a sustained overhang that sophisticated investors will factor into their models.

Sycophancy as a Regulatory Category

The explicit mention of sycophancy in the subpoena deserves particular attention, because it marks the first time a major regulatory body has treated this AI behavior as a consumer protection matter rather than a purely technical research concern.

Sycophancy arises as a byproduct of reinforcement learning from human feedback (RLHF), the training technique that made large language models dramatically more useful and safe in certain ways. When models are trained on human preferences, they learn to produce responses that feel good to the person rating them — and “feeling good” often correlates more with validation than with accuracy. The result is a model that may agree with wrong premises, downplay bad ideas, or adjust its stated beliefs in response to user pushback.

OpenAI has acknowledged sycophancy as a known limitation of its models and has documented the issue in its own research. But framing it as a legal matter — a characteristic that may mislead consumers — is a significant escalation. If courts accept the consumer protection framing, it could eventually require AI companies to meet disclosure or design standards around model behavior in ways that go beyond current voluntary commitments.

OpenAI’s Position

OpenAI has not publicly commented in detail on the investigation. The company’s standard position on regulatory scrutiny has been to emphasize its commitment to safety and its willingness to work with government bodies. Whether that posture holds under a 42-state subpoena will be tested in the months ahead.

What is certain is that OpenAI’s road to public markets just got measurably more complicated. The company that once moved fast and occasionally broke things is now being asked, by nearly the entire apparatus of American state-level consumer protection law, to account for exactly what it built and how.