The complaint opens with a quotation, not a statute. “AI will probably most likely lead to the end of the world, but in the meantime, there’ll be great companies.” Texas Attorney General Ken Paxton attributes the line to Sam Altman, from 2015, and uses it to establish what the state says OpenAI knew about its own product before it sold the product to Texans.
Paxton filed the suit on Oct. 23 in Collin County district court against OpenAI Global LLC, several affiliated entities and Altman as an individual. The claims rest on two Texas statutes: the Deceptive Trade Practices and Consumer Protection Act, which governs how goods and services may be marketed, and the Securing Children Online through Parental Empowerment Act, which requires age verification and parental consent on platforms children use.
The theory is narrow enough to survive a first motion to dismiss. Texas does not argue that ChatGPT produces wrong answers, which would invite an argument about the state of the technology. It argues that OpenAI marketed the system as safe, accurate and reliable while knowing it could produce harmful output, and that the gap between the marketing and the knowledge is the violation. Children are named as a category of user exposed to that gap, without the age checks the SCOPE Act requires.
The requested remedies are where the case gets unusual. Texas wants civil penalties of up to $10,000 for each violation, surrender of profits the state characterizes as unlawfully obtained, and an injunction with two specific terms: that OpenAI block minors from the product, and that a court-appointed evaluator sign off on new models before release. The second ask has no close parallel in consumer protection litigation. Courts have appointed monitors to oversee banks and unions. Appointing one to clear a machine-learning release schedule would put a judge’s expert inside the product cycle of a company that has never been audited quite this way.
Paxton’s public statement was written for a press release, and reads like one. His office said mass shooters had been aided in deadly attacks, that vulnerable people had been encouraged into suicide, that professionals had suffered public humiliation, and that minors had become addicted to a tool that “fakes human compassion to collect their data with no parental oversight.” The complaint itself starts wider, noting that artificial intelligence now accounts for more than a third of total U.S. market value, a figure the state uses to argue that the gains were not earned.
The SCOPE Act gives the second claim its teeth and its weakness. Texas passed it in 2023 and it took effect in 2024, requiring platforms to verify the age of users and obtain parental consent for minors. Trade groups challenged it in federal court, and parts of the law have been paused while the case proceeds. A state suing under a statute that is itself under injunction is a familiar posture in Austin, but it means the children’s claim may move more slowly than the marketing claim.
The Deceptive Trade Practices Act is older and better tested. Enacted in 1973, it lets both consumers and the attorney general sue over false or misleading representations, and its penalty structure escalates when the victim is elderly. Texas has used it against insurers, car dealers and, more recently, technology platforms. What it has not often done is apply it to a company whose product output varies by prompt, which is the defense OpenAI is likely to build: a system that can be wrong is not a system that was misrepresented, provided the possibility of error was disclosed.
OpenAI had not issued a public response as of Monday, and the company has said nothing publicly about the specific allegations. Its likely defense is visible in the complaint’s own framing, which turns on what the company knew and when, rather than on whether any single answer was wrong.
Naming Altman personally is a choice with consequences beyond symbolism. It exposes the chief executive to discovery about internal decisions, including which risks were measured, which were escalated and which were left in a slide deck. Consumer protection suits against corporations tend to end in settlements funded by insurance. Suits against an individual officer tend to produce depositions that are harder to seal.
The political context is not incidental. Paxton is running for the U.S. Senate and has built a national profile on suing large technology companies, often in venues where the filing itself is the message. State attorneys general have become the most active AI regulators in the country while Congress has passed no general statute, and a Texas action gives counterparts in other states a template to copy with their own consumer laws.
For the industry, the practical exposure sits in marketing copy, model cards and help-center pages. Claims that a system is safe, that it refuses harmful requests, or that it is suitable for children are now potential exhibits. Companies that write those pages will start routing them through counsel, and product teams will find that the sentence describing what a model cannot do is the one that gets deleted.
The most durable output of the case may be paper. Even if the injunction is trimmed on appeal, discovery will produce internal assessments of harm, and those documents will be cited in the private suits that follow. Texas has bought itself a seat at the table where the industry’s safety claims get audited, and it did so for the price of a filing fee.
