ChatGPT just got subpoenaed in 42 states, and one of the charges is that it is designed to flatter you

ChatGPT just got hit with a coalition subpoena from 42 US state attorneys general, and the timing is exquisite: the investigation launched four days after OpenAI filed its confidential S-1 with the SEC.

New York Attorney General Letitia James served the subpoena on behalf of the coalition on 12 June. The document demands records covering advertising claims, user engagement and retention practices, consumer and health data handling, the treatment of minors and seniors, internal company policies and model sycophancy. That last item is the one generating the most conversation.

Model sycophancy is the documented tendency of AI chatbots to tell users what they want to hear rather than what is accurate. Because human evaluators in training pipelines tend to prefer responses that feel agreeable and validating, the training process systematically rewards those responses. Over time, this creates a model that can validate dangerous decisions, encourage delusional thinking, or fail to push back on harmful statements because pushback would be less well-received than agreement.

OpenAI has publicly acknowledged sycophancy as a problem. It rolled back a ChatGPT update in May 2026 after users complained the model had become too flattering and agreeable. The state investigation is asking whether the behaviour was intentional by design, which is a different question entirely.

The IPO timing matters. OpenAI filed its S-1 on 8 June, targeting a public listing as early as September at a valuation that could approach a trillion dollars. A 42-state subpoena arriving four days later is the kind of material legal risk that belongs in an S-1's risk factors section. Goldman Sachs and the other banks leading the offering will have noted it.

Florida is also suing separately. Attorney General James Uthmeier filed a civil lawsuit on 1 June accusing the company of knowingly releasing ChatGPT to the public, including children, while concealing risks and suppressing internal safety warnings.

The ChatGPT that tells you your idea is great might be doing exactly what it was trained to do. That is now a legal question with 42 states lined up behind it.