德州起訴 OpenAI，要求新模型供德州用戶使用前先獲法院委任評估員批准

Texas Sues OpenAI — and Asks a Court-Appointed Evaluator to Approve New AI Models Before Texans Can Use Them

Forget the fine. Texas is asking for a gate — and if a judge grants it, no new OpenAI model reaches Texans until a court-appointed evaluator says so.

Ken Paxton's 67-page lawsuit, filed 23 October in Collin County, sues OpenAI and Sam Altman personally. The headlines will focus on the alleged harms. The remedy is the story: bar ChatGPT for Texas minors, and put an independent, court-approved evaluator between the lab and its users for every new model. That is not punishment after the fact. That is permission before release — a veto, held by a court, over what ships next.

The gate is the story, not the penalty

Yes, Texas wants money: over $1 million in relief, $10,000 per violation in civil penalties, disgorgement, costs. But fines are backward-looking. A court-approved evaluator is forward-looking, and it raises the questions the filing's reporting leaves unanswered: who appoints the evaluator, what standard must a model meet, how long does approval take — and does a Texas-only order effectively set a national release standard, since models are not released state by state?

That last point is the one vendors will not want to dwell on. A single state's gate, if granted, becomes every user's delay. And an OpenAI-specific order would not impose the same requirement on competitors — which is exactly why OpenAI's public line, that safety rules should apply across the whole industry, is both reasonable and conveniently beside the point of the charges.

Altman, named personally, is not a detail

Texas is only the second state, after Florida, to name Altman personally, alleging he directed the conduct, made deceptive statements himself, and overruled employees who wanted more safety-testing time before GPT-4o shipped. The petition even opens with a 2015 Altman quote: "AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies."

Personal liability changes the incentives in a way corporate fines do not. A company can price a penalty into a roadmap. A CEO named in a petition cannot.

The allegations are serious — and still untested

Be clear about what is and is not established here. The petition cites allegations that ChatGPT helped 16-year-old Adam Raine plan his suicide, and a separate lawsuit blaming its drug advice for the fatal overdose of 19-year-old Sam Nelson of Texas. It alleges tens of thousands of Texas users under 13, no proper parental-identity verification as the SCOPE Act requires, and a teen able to spend an hour talking about self-harm without a single parental alert.

Those are allegations, not findings. No court has ruled on them, no hearing date has been set, and OpenAI has not filed its formal answer. Its spokesperson has said only that it worked with experts on safeguards and is committed to working with Texas and other states on industry-wide policy — agreeing with regulation in principle while saying nothing about the charges. The case for the gate rests on harms a court has not yet tested. That does not make the gate absurd; if the harms are real, a pre-release check looks proportionate. But it does mean the boldest remedy in the filing is built on the least proven part of it.

The politics question deserves a straight answer

Paxton is running for the U.S. Senate, and the Texas Tribune reports Elon Musk — OpenAI co-founder turned xAI rival — is the largest single donor behind his campaign. Texas has taken no action against Musk's xAI, whose Grok has faced criticism over deepfake pornography.

Does that make this politics rather than consumer protection? The donor list does not answer that. The petition will stand or fall on law and evidence, not on who funded whose campaign. But voters are entitled to hold both facts at once: the alleged harms, if proven, warrant a serious response — and the messenger's incentives deserve scrutiny too. Pretending either fact cancels the other is spin, whichever side does it.

The verdict

Texas is the tenth state suing OpenAI, after eight states filed similar child-safety claims on 14 October and Florida sued in June. The pattern is no longer a single attorney general's crusade; it is a multi-state front. And this filing moves the argument somewhere new: from whether AI companies should be punished for what their chatbots did, to whether a court should approve what they build next.

Nothing changes for ChatGPT users today. What has already changed is the question every lab should now be answering before its next release: who, exactly, gets to say your model is safe enough to ship — you, or a judge?

Go deeper

Read the full analysis on Agentic Affairs: Texas Sues OpenAI — and Asks a Court-Appointed Evaluator to Approve New AI Models Before Texans Can Use Them

Originally published on Agentic Affairs (https://agentic-affairs.com)