42 Attorneys General Subpoenaed OpenAI's Ads, Its Safety Policies, and Its Sycophancy All at Once

There's a specific kind of institutional reckoning that announces itself not with a bang but with a subpoena. Forty-two state attorneys general have opened a sweeping investigation into OpenAI — and the breadth of what they're asking for is the actual story.

According to Tom's Hardware, the subpoena targets ads, user retention practices, data handling, minors, health data, model behavior, and safety policies. TechCrunch notes the questions extend to OpenAI's ad policies and its handling of health data specifically. That's not a narrow probe looking for a clean violation. That's a coalition of regulators essentially asking: walk us through everything you do and explain why any of it is acceptable.

When investigators subpoena your sycophancy — the documented tendency of AI models to tell users what they want to hear rather than what's accurate — you've crossed from product liability into something more philosophical. The government is now formally curious about whether your software is manipulating people by design.

The Business Model Under the Microscope

What makes this investigation structurally different from past tech scrutiny is that it arrives at an unusually exposed moment. Tom's Hardware reported the probe came just days after OpenAI's reported IPO filing. That timing matters. An IPO is when a company has to stand behind its numbers, its practices, and its story in a way that press releases never require. Regulators know this. Filing to go public while 42 state AGs are circling your data practices is not a comfortable position to be in.

The scope of the subpoena suggests the investigators aren't looking for a single bad actor moment — a rogue ad sale, one mishandled health record. They're building a picture of the whole operation. Ads. Retention. Minors. Safety. Model behavior. These aren't different problems. They're facets of one question: does the way OpenAI makes money conflict with the way it represents itself to the public?

That's the question the industry has been successfully dodging for years, wrapped in the language of responsible development and safety commitments. It turns out the language doesn't satisfy 42 prosecutors.

We've Seen This Before, and It Still Matters

The cynical read is that this is regulatory theater — a coordinated show of force that produces a settlement, a few policy tweaks, and then business as usual. We've watched that movie with social media platforms. We know how it ends.

But there's something meaningfully different here: the specific inclusion of model sycophancy as an investigative target. That's not a privacy violation or a terms-of-service breach. That's regulators asking whether the AI itself is a manipulative actor — whether the product's core behavior is designed to keep users engaged at the cost of their actual interests. No platform has faced that question in quite this form before.

If the investigation forces real documentation of how these systems are tuned, what the retention incentives are, how health data flows, and how minors are treated — that's not nothing. That's the kind of paper trail that rewrites how an industry operates.

Forty-two states just put the whole business model in the witness box. The interesting part isn't whether OpenAI survives. It's what the answers look like under oath.