Opinion: Suing OpenAI Is Not the Same as Regulating AI

Eight state attorneys general sued OpenAI on the same day last month, each in a separate courthouse, each alleging the same essential thing: that the company marketed ChatGPT as safe for children while building a product that, by its own internal admissions, was not. Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed on October 14. Texas followed on October 23, with Attorney General Ken Paxton naming Sam Altman personally as a defendant alongside the company. Florida had already done the same in June. Ten states, two methods of attack, one defendant: a company now seeking a public valuation that bankers have put as high as $1 trillion.

This is not a stray legal nuisance for OpenAI. It is the first sustained attempt by American government, at any level, to use the law rather than a stern letter to constrain how a frontier artificial-intelligence company builds and sells its product. That is worth taking seriously on its own terms, independent of whether any single count survives a motion to dismiss. It is also worth being skeptical of, because the pattern of who is suing, whom they are leaving alone, and why reveals less a coherent regulatory strategy than a scramble by state officials to occupy ground that Washington has left empty.

What the states are actually alleging

The petition Iowa filed in Polk County District Court is not a thin document. It runs to roughly ninety pages and reads less like a product-liability brief than an indictment of a company's culture. It alleges that OpenAI rushed ChatGPT to market in November 2022 ahead of rivals that had built comparable systems but held them back over safety concerns, that the company's own former chief technology officer and head of alignment raised internal alarms, and that the board fired Altman in 2023 for not being, in its own words, "consistently candid." It cites OpenAI's own disclosures that roughly a million people each week discuss suicide in conversations with ChatGPT, and it catalogs a string of lawsuits already filed by grieving families, including the 2025 wrongful-death suit brought by Matthew and Maria Raine over the death of their 16-year-old son, Adam, in San Francisco Superior Court.

Iowa is not asking a jury to decide whether ChatGPT caused any particular death. It is asking a court to find that OpenAI violated the state's Consumer Fraud Act by misrepresenting the product's safety, and it wants civil penalties of up to $40,000 per violation, plus disgorgement of revenue earned in the state. Texas wants something structurally more ambitious: an injunction that would bar minors from the product and require a court-appointed evaluator to sign off on future model releases before they reach Texans. Both theories rest on the same underlying claim, one the company disputes: that a product used by roughly half of American adults, by Pew Research Center's count from its February 2026 survey, with 24 percent using chatbots daily, was deployed faster than it was made safe.

“Because of Defendants’ repeated misrepresentations and insatiable hunger to win the AI race—even if it means recklessly introducing unsafe ChatGPT models to the public—mass shooters have been aided and abetted, vulnerable citizens have been coached on suicide…”

That line, from Iowa's petition, is deliberately prosecutorial, and it should be read as advocacy, not adjudicated fact. OpenAI has not been found liable for anything in these cases, and the company has pointed to real changes it has made, including an age-prediction system it rolled out in January that defaults unverified accounts to a restricted teen experience. A spokesman told reporters after the October filings that the company has built age-specific safeguards competitors lack and that no system can be made perfect. Both things can be true: that OpenAI has built real guardrails, and that it built them only after years of a product that, per the lawsuits, lacked basic ones.

A regulator vacuum, filled unevenly

The reason ten attorneys general are doing this work with nineteenth-century consumer-fraud statutes is that nobody else is doing it with anything newer. The Federal Trade Commission's own September 2025 orders to OpenAI and six rivals, issued unanimously under Section 6(b) of the FTC Act, were explicitly a study, not an enforcement action; the agency said so itself. Congress has passed nothing resembling a federal AI-safety statute. Into that vacuum has stepped a patchwork of state officials, each empowered only within their own borders, each applying statutes written for used-car dealers and timeshare pitches to a technology that did not exist when those laws were drafted.

That patchwork is not neutral. Texas's own record on AI harms to children is selective in a way that undercuts the moral clarity of its filing. Elon Musk's xAI built Grok, a rival chatbot that California's attorney general, Rob Bonta, found serious enough to warrant a cease-and-desist letter in January over Grok's generation of sexualized images of real women and children. Texas has brought no comparable action against xAI. Musk, meanwhile, is reported to be the single largest donor to Paxton's Senate campaign, having contributed close to $200 million. None of that proves Paxton's OpenAI suit is pretextual; the underlying allegations against ChatGPT, many lifted nearly verbatim from the coordinated state filings, are serious enough to answer regardless of who is asking. But a legal strategy that happens to spare a megadonor's product while targeting a chief rival's is not the kind of evenhanded enforcement that builds public confidence in the project of regulating AI at all.

Litigation is a poor substitute for a rulebook

There is also a harder problem underneath the politics. Consumer-fraud litigation is built to punish specific misrepresentations after the fact. It is not built to set forward-looking standards for how a model should be tested before release, what counts as adequate crisis-intervention design, or how a company should weigh engagement against risk when the two are, as several of these complaints argue, commercially aligned. A $40,000 civil penalty is a rounding error against a company reportedly generating more than $25 billion in annualized revenue. Even the more ambitious Texas remedy, third-party sign-off on new models, would apply only within Texas, creating the possibility of fifty different safety regimes for a product that does not respect state lines.

None of this is an argument against the lawsuits. Suing is what officials do when legislatures will not legislate, and the factual record these filings compile, drawn from OpenAI's own statements and from court files in wrongful-death cases already underway, deserves scrutiny whatever a judge eventually decides about deceptive-trade-practices law. But a country that governs the most consequential consumer technology of the decade through a dozen uncoordinated state fraud suits, filed by officials with their own incentives and blind spots, has not solved its AI-safety problem. It has merely found the only lever available while Congress and the FTC continue to defer the question of writing an actual rule.