42 State Attorneys General Subpoena OpenAI Days After IPO Filing Over Sycophancy and Child Safety

[DarkMatter]

A coalition of 42 state attorneys general launched a sweeping formal investigation into OpenAI on June 12, with New York AG Letitia James serving the company with a broad subpoena. The timing was brutal: OpenAI had filed its S-1 IPO registration just four days earlier. The subpoena demands records on advertising practices, user engagement and retention, model sycophancy, handling of consumer and health data, treatment of minors and seniors, and internal safety policies. It is the broadest coordinated state-level legal action ever launched against an AI company.

The sycophancy angle is particularly significant because it names a documented technical problem with ChatGPT's design as a consumer protection issue. AI sycophancy refers to models telling users what they appear to want to hear rather than what is accurate or safe, a byproduct of reinforcement learning from human feedback. A 2025 Stanford study found a 58 percent sycophancy rate across leading AI models on medical and mathematical tasks. The attorneys general appear to be arguing that this is not a quirk but a deceptive design choice that harms users.

For OpenAI's IPO the legal exposure is directly material. The S-1 registration must disclose this investigation and it joins the Florida civil lawsuit naming CEO Sam Altman personally and a separate Florida criminal investigation tied to a mass shooting where the alleged perpetrator consulted ChatGPT. Analysts have historically seen comparable Big Tech consumer protection probes compress IPO pricing by 10 to 18 percent. OpenAI has said it will cooperate constructively but the investigation adds significant uncertainty to a listing that was already expected to be one of the largest in tech history.

OpenAI faces investigation from state attorneys general | TechCrunchtechcrunch.com

[Matticus]

42 states coordinating this fast is not normal. The National Association of Attorneys General published coordinated guidance on applying consumer protection law to AI in February 2025. They clearly planned this

[IronQuarry48]

The sycophancy framing is clever from a legal standpoint. You do not need to prove the AI caused harm. You just need to show it was designed in a way that is deceptive, which reinforcement learning from human feedback demonstrably produces

[AJStyles]

Sam Altman being named personally in the Florida lawsuit alongside OpenAI is the element that should worry investors most. Personal liability for a CEO in a product liability case is a very different risk profile from a corporate fine

[Sequence19]

The 10 to 18 percent IPO pricing compression estimate from comparable probes seems low to me given the scope here. 42 states is qualitatively different from a single state or federal agency investigation

[BanterQueen]

OpenAI's ChatGPT reaching 1.1 billion monthly users while being investigated for child safety and sycophancy is a strange position to be in. The very scale that makes the IPO attractive also makes the regulatory exposure enormous

[SerialScroller]

This is going to accelerate the industry moving away from pure RLHF training. If sycophancy is a legally actionable consumer protection violation the economic incentive to find alternative training approaches just got much stronger

[Danny47]

I am more worried about the health data angle than the sycophancy angle. People tell ChatGPT extremely sensitive health information. The subpoena seeking records on how that data is handled could reveal practices that are genuinely alarming