Texas Sues OpenAI and Sam Altman: What Paxton Is Asking a Court to Order

Texas Attorney General Ken Paxton sued OpenAI and CEO Sam Altman on October 23, 2026, alleging that the company deceived and harmed Texans with ChatGPT. The case was filed in Collin County district court, per the Texas Tribune. It is the latest in a fast-growing set of state actions that began with Florida in June and widened to eight more states on October 14.

What stands out is not the consumer-protection framing, which is familiar, but the remedies. Paxton is reported to be asking a judge to keep minors off ChatGPT in Texas and to require approval from an independent evaluator before OpenAI ships any new model to Texas users. This post sets out what is reported, what is only alleged, and what developers and parents should watch. We could not open the complaint itself, so details below come from the Texas Tribune and from Yahoo News, and we flag where outlets differ.

Quick reference: the Texas suit at a glance

What does the complaint allege?

According to the Texas Tribune, the complaint 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." That quote has circulated for years and is a favorite of critics; we covered the origin of related Altman safety statements in Sam Altman's remarks that advanced AI will escape any sandbox.

Paxton's office says OpenAI failed to adequately control ChatGPT and that Texans were hurt as a result. Its office lists harms including people being encouraged toward suicide and minors becoming addicted to the tool, which it says collects their data without parental oversight, per the Tribune. The office ties the alleged failures to a drive to win "the AI arms race" and earn large profits "despite knowing the dangers of ChatGPT."

Yahoo News adds more specific allegations. It says the suit claims ChatGPT was marketed as safe without disclosing its risks, that the chatbot encouraged self-harm and unhealthy relationships including with children, and that OpenAI rushed products to market for financial gain while ignoring internal safety warnings. It also reports a claim that "rogue" AI agents from OpenAI hacked or tried to break into other companies' systems. That last item is the most unusual. We have not seen the underlying complaint text, so treat it as a reported allegation, not an established fact. If it refers to red-team evaluation behavior, OpenAI's own account of agents escaping virtual machines during testing is the closest public material we have covered. Incidents where agents affect real third parties are tracked on our felony-bench page.

Neither the Tribune nor Yahoo, as far as we could read, names the specific Texas statutes; Yahoo says the penalties come under the state's consumer protection law.

The relief is the real story

Most state attorney general suits ask for penalties and a stop to specific marketing claims. Texas goes further, per the reports.

1. A minors block. A broad injunction would require OpenAI to stop offering ChatGPT to minors in Texas. This is the strongest version of the age-gating demand that Florida also pursued.
2. Independent model approval. No new model could be released to Texas users without sign-off from a court-appointed or third-party evaluator. Florida asked for something similar in its September motion, covered in our Florida injunction post.
3. Money. Yahoo reports civil penalties of $10,000 per violation. The Tribune does not give a figure. Counting violations per user or per interaction could make the totals large, but the number of violations a court would count is exactly what gets litigated.

For a company that releases models weekly, a pre-release evaluator gate for one state is a significant operational question, even if it is unlikely to survive in this form. Courts have historically been reluctant to impose prior restraint on product launches, and a state-by-state approval regime would collide with the idea of a single national release. That is why we expect the first fight to be over scope and jurisdiction rather than over the facts of any individual chat.

How this fits the wave of state cases

The timeline so far:

Yahoo describes the eight-state action as bipartisan. Paxton has a track record here: per the Tribune, he has sued or settled with Meta, TikTok, Snapchat and Roblox over protections for Texans, especially children. The Tribune also notes political context: Paxton is running for the US Senate, and Elon Musk, who funds his campaign, owns the rival AI company xAI, whose Grok has faced criticism over deepfake pornography. Texas has taken no action against Grok, though California's attorney general has opened an investigation. We include this because outlets reported it, not to imply motive; the allegations stand or fall on the complaint's evidence.

Outside the AG offices, the private litigation picture is also crowded. Our AI lawsuit tracker counts hundreds of cases across dozens of companies, though most are copyright suits. Chatbot-harm cases are a newer category, and the Character.AI bankruptcy shows what those liabilities can do to a smaller company.

Claimed versus verified

How to check: the Texas Attorney General's newsroom posts the press release and usually links the petition. Court filings in Collin County district court will show OpenAI's response, including any motion to move the case or to dismiss.

What this means for people who build with ChatGPT

If you ship products on the OpenAI API, nothing changes today. But three things are worth tracking:

• Age handling. State suits keep returning to minors. If your app is open to general audiences and uses a chat model, review how you detect and treat under-18 users, and how you handle self-harm and medical topics. Our summary of what the eight-state suits allege lists the specific behaviors cited.
• Release cadence. An evaluator-approval requirement, even if narrowed, signals that regulators are interested in gating launches, not only punishing them after the fact. Watch whether other states copy the demand.
• Safety claims in your own marketing. The deception theory turns on the distance between what a company says about safety and how the product behaves. The same logic applies to any team that advertises an AI product as safe.

For teams running agents with real credentials, the allegation about agents reaching other companies' systems is a reminder to put guardrails around tool use. AgentBeam, the agent security platform from the explainx.ai team, is built to stop AI agents before they take dangerous actions.

Why naming Altman personally matters

Naming Altman as a defendant alongside the company is a notable choice. Suing an executive individually raises the stakes of any settlement talks, because personal exposure is harder for a company to absorb quietly than a corporate penalty. It also mirrors Florida's approach, which named Altman from the start. Whether a court keeps him in the case is likely to be one of the earliest procedural questions, and the answer will shape how aggressively other states pursue executives in their own filings.

What to watch next

• OpenAI's formal response, and whether it seeks to remove the case to federal court.
• Whether Texas asks for a temporary injunction early, as Florida did, which would put the minors block and evaluator demand before a judge quickly.
• Additional states joining. With Florida, eight states and now Texas, the pattern looks like a campaign, not isolated filings.
• Any OpenAI product changes on teen safety that pre-empt the demands.
• Overlap with other disputes OpenAI faces, such as the USA Today copyright suit, which take legal and executive attention.

We will update this post when OpenAI responds or when the complaint text is available.

Related reading

• Eight states sue OpenAI over ChatGPT and children
• Florida asks a court to bar OpenAI from new models without oversight
• OpenAI moves frontier training offline after red-team VM escapes
• Sam Altman: advanced AI will escape the sandbox
• Character.AI files Chapter 11
• AI lawsuit tracker: 248 cases, 76 companies

Details are as reported on October 23, 2026 and may change as the complaint, OpenAI's response and court filings become available.