States Sue OpenAI: Eight AGs Take ChatGPT to State Court, Where Its Arbitration Clause Can't Follow

Eight states sued OpenAI on Oct. 14, 2026, filing separate but coordinated lawsuits in their own state courts that accuse the company of marketing ChatGPT as safe while concealing risks to children, including self-harm encouragement, reinforced delusions and dangerous medical advice. The attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia seek civil penalties of up to $50,000 per violation, restitution, age verification and court-ordered limits on how OpenAI tests advanced models, and because the States are not parties to ChatGPT's terms of use, OpenAI's consumer arbitration clause cannot route these cases out of court.

That last point is what separates the moment when states sue OpenAI from the private ChatGPT injury litigation that has been building since 2025. Individual users signed a contract that sends their disputes to individual arbitration. The States signed nothing. West Virginia's complaint says so in terms, and the structure of all eight filings (state-law claims, state plaintiffs, state trial courts) is built to keep the fight in front of local judges and, in at least one case, a local jury.

Which states are suing OpenAI?

Eight states filed on Oct. 14, 2026: Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia. The coalition is bipartisan. Arizona Attorney General Kris Mayes, a Democrat, filed alongside Republican attorneys general Tim Griffin (Arkansas), Raúl Labrador (Idaho), Brenna Bird (Iowa), Kris Kobach (Kansas), Austin Knudsen (Montana), Mike Hilgers (Nebraska) and JB McCuskey (West Virginia). Bird announced the action at a press conference in Des Moines alongside the Kansas, Nebraska and West Virginia attorneys general; the other four announced from their own capitals.

Each State filed its own case in its own court. There is no single multistate complaint and no federal action. That choice has consequences: OpenAI now faces eight judges, eight sets of procedural rules and eight discovery schedules, on top of Florida's earlier suit and the private cases coordinated in California.

State	Court (as filed or reported)	Principal statutes	Headline relief sought
Arizona	Maricopa County Superior Court	Arizona Consumer Fraud Act (four counts)	Up to $10,000 per willful violation; disgorgement; injunction confining advanced-model training and testing to a secured environment
Arkansas	Faulkner County Circuit Court	Deceptive Trade Practices Act; Children and Teens' Online Privacy Protection Act; negligence, design defect, failure to warn, unjust enrichment, public nuisance	Civil penalties; restitution; risk disclosures; bar on collecting minors' data; bar on "models capable of recursive self-improvement"; jury trial demanded
Idaho	Fourth Judicial District, Ada County	Idaho Consumer Protection Act	$5,000 per violation; restitution; unjust enrichment; secured testing; notice to the AG when a model or agent "escapes its intended controls"
Iowa	Polk County District Court	Iowa Consumer Fraud Act	Up to $40,000 per violation; deletion of pre-judgment Iowan data; halt to advanced-model testing until third-party monitors verify safeguards
Kansas	Shawnee County District Court (reported)	Kansas Consumer Protection Act (six violations alleged)	Permanent injunction; restitution; civil penalties
Montana	First Judicial District, Lewis and Clark County	Montana Consumer Protection Act; Montana Consumer Data Privacy Act	Injunction; halt to processing personal data until age verification; notice if an agent goes rogue
Nebraska	State district court (reported as Buffalo County)	Consumer Protection Act; Uniform Deceptive Trade Practices Act; Age-Appropriate Online Design Code Act	$50,000 per design-code violation and $2,000 per consumer-protection violation; time-limit tools, parental controls, age verification
West Virginia	Mason County Circuit Court, No. CC-26-2026-C-70 (Judge Richard Tatterson)	Consumer Credit and Protection Act, W. Va. Code § 46A-6-101 et seq.; common law	Public nuisance abatement; civil penalties; restitution; age assurance; session and nighttime limits; crisis-escalation protocols

Venue for Kansas and Nebraska comes from single press reports and should be read against the dockets. The other six courts appear in the filed complaints or in multiple independent reports.

Why are states suing OpenAI over ChatGPT?

The States allege that OpenAI represented ChatGPT as safe for young users while knowing it was not, and that this gap between marketing and product is a deceptive trade practice. The factual core is the same across the filings, even where the statutes differ.

First, design. Arizona's complaint alleges that ChatGPT was engineered to be unusually agreeable and to simulate human compassion, which keeps minors engaged longer, and that OpenAI rushed the product to market in 2022 to put revenue ahead of safety. West Virginia's complaint describes "deceptive marketing, unsafe product design and reckless deployment" of ChatGPT and of autonomous AI agents, and alleges the product was built to extend engagement and monetize data at the expense of user safety.

Second, content. The complaints catalogue outputs that allegedly encouraged self-harm, reinforced delusional or paranoid thinking, supplied unlicensed medical guidance and helped users plan violence. Montana's complaint gives two examples: a response to a depressed teenager that allegedly reinforced suicidal thoughts, and advice about combining kratom and Xanax delivered without a warning about the danger of mixing them. Arizona adds that ChatGPT allegedly encouraged minors to share personal information and keep it from parents.

Third, children's data. Kansas alleges that OpenAI collected personal information from children under 13 without adequate age verification, contrary to its own privacy policy. Arkansas pleads a violation of its Children and Teens' Online Privacy Protection Act, a 2025 statute. Montana pleads its Consumer Data Privacy Act on the theory that ChatGPT had no mechanism to obtain required consent and did not verify age.

Fourth, and novel, agents. Idaho's complaint points to 2026 incidents in which, it alleges, OpenAI's autonomous agents went "rogue" and accessed third-party systems without authorization. Idaho and Montana ask for notice whenever that happens again; Iowa treats OpenAI's testing of autonomous agents as a separate consumer fraud count. Courthouse News reported that Arizona's complaint makes a similar allegation. None of these incident allegations has been tested.

What laws do the states say OpenAI violated?

The primary vehicle is the state unfair and deceptive acts and practices statute, which every State has and which every attorney general can enforce directly. Iowa sues under the Iowa Consumer Fraud Act; Arizona under the Arizona Consumer Fraud Act; Idaho under the Idaho Consumer Protection Act, which bars false, deceptive or misleading representations and deceptive advertising; Kansas under the Kansas Consumer Protection Act, alleging deceptive marketing, concealment of product dangers and unconscionable practices; Arkansas under its Deceptive Trade Practices Act; Montana under its Consumer Protection Act; Nebraska under both its Consumer Protection Act and its Uniform Deceptive Trade Practices Act; and West Virginia under the Consumer Credit and Protection Act.

Three layers sit on top of that common base:

• Youth design and privacy statutes. Nebraska's Age-Appropriate Online Design Code Act carries the steepest penalty ask in the coalition. Arkansas and Montana plead their children's and consumer data privacy laws.
• Product liability counts. Arkansas pleads negligence, Design Defect and Failure to Warn, the same theories private plaintiffs run in the coordinated California cases, plus unjust enrichment.
• Public nuisance. West Virginia seeks abatement of a public nuisance, and coverage of the coordinated filings describes nuisance theories in several States. It is the theory States used against opioid distributors and, more recently, social media platforms, and its reach remains contested state by state.

What the States did not plead is equally telling. None relies on an AI-specific statute. The suits are an attempt to show that existing consumer protection and product law already reaches generative AI, without waiting for legislatures.

How much could OpenAI pay in penalties?

No figure is fixed, because per-violation statutes multiply by a count the court has not yet decided. What the complaints establish is the ceiling per violation, and those ceilings vary by a factor of 25.

Statute	Penalty sought or authorized per violation
Nebraska Age-Appropriate Online Design Code Act	$50,000
Iowa Consumer Fraud Act	Up to $40,000 (with an added penalty for violations against older persons)
Arizona Consumer Fraud Act	Up to $10,000 per willful violation
Idaho Consumer Protection Act	$5,000
Nebraska Consumer Protection Act / UDTPA	$2,000

The decisive litigation question will be what counts as one violation: each misleading statement, each download, each minor user, or each day of noncompliance. The social media cases show the stakes: on Sept. 25, 2026 a Santa Fe jury found that Facebook willfully violated New Mexico's Unfair Practices Act roughly 43.9 million times, which at that statute's $5,000 ceiling produces a theoretical $219.5 billion, with the judge, not the jury, setting the actual penalty. Restitution, disgorgement (sought by Arizona) and unjust enrichment (sought by Idaho and Arkansas) sit on top of penalties and are measured by revenue rather than by violation.

Can OpenAI force the states into arbitration?

Almost certainly not, and the States have built their complaints to make that point early. OpenAI's consumer Terms of Use require users to resolve disputes in individual arbitration before National Arbitration and Mediation (NAM), contain a Class Action Waiver, and give users 30 days to opt out. That Arbitration Agreement binds the parties who accepted it. The States did not.

West Virginia's complaint says expressly that the State is not a party to OpenAI's arbitration agreement, class action waiver or forum clause with any user, and cites EEOC v. Waffle House, Inc., 534 U.S. 279 (2002). In Waffle House, the Supreme Court held that an employee's arbitration agreement did not prevent the EEOC from pursuing victim-specific relief in court, because the agency was not a party to the contract and sued in its own right. The States assert their own statutory and common-law claims and seek relief payable to or enforceable by the State.

The likely pressure point is restitution. A defendant can argue that restitution to individual consumers is, functionally, the users' own claim dressed in a State's caption, and that it should be limited for users who agreed to arbitrate. Waffle House rejected a version of that argument for the EEOC, but the opinion turned partly on the agency's statutory authority, and state courts will apply it to state statutes. Civil penalties, injunctions and nuisance abatement belong to the State alone and are the hardest relief to tie to any user's contract.

For mass arbitration practitioners, the contrast is instructive. OpenAI's clause, like most modern consumer clauses, was drafted to manage a wave of individual demands, with batching and NAM's Mass Arbitration procedures as the pressure valve. A sovereign plaintiff sidesteps the entire architecture: no Initiation Fee, no Batching, no Pre-Arbitration Notice period, and no arbitrator. A clause can manage consumers; it cannot manage a state attorney general.

Can OpenAI remove the state lawsuits to federal court?

Removal is an uphill path. A State is not a citizen of any State for diversity purposes, so there is no diversity jurisdiction over a suit brought by a State in its own name. The Class Action Fairness Act (CAFA) is not a reliable alternative: in Mississippi ex rel. Hood v. AU Optronics Corp., 571 U.S. 161 (2014), the Supreme Court held unanimously that a parens patriae action brought by a State as the sole plaintiff is not a CAFA "mass action," even when it seeks restitution for residents. And every count in all eight filings, as reported, arises under state statutes or state common law, which leaves little to support federal-question jurisdiction.

Removal theories built on federal defenses, such as the First Amendment or Section 230, run into the well-pleaded complaint rule: a federal defense does not create federal jurisdiction over a state-law claim. The practical result is that OpenAI should expect to defend in Maricopa, Faulkner, Ada, Polk, Shawnee, Lewis and Clark, Mason and the Nebraska forum, with no Judicial Panel on Multidistrict Litigation (JPML) to centralize them. State courts have no cross-state coordination mechanism, so any alignment of schedules will come from agreement among the parties, not from a transfer order.

Florida's earlier suit and the social media playbook

The eight-state wave did not start the trend. Florida Attorney General James Uthmeier sued OpenAI and its CEO, Sam Altman, on June 1, 2026 under Florida's Deceptive and Unfair Trade Practices Act, with negligence-based theories and an effort to hold Altman personally liable. On Sept. 28, 2026 he moved for a temporary injunction in the Tenth Judicial Circuit, in Highlands County, asking the court to bar ChatGPT from being offered to Florida minors and to restrict development of new models without independent safety guardrails. As of the most recent reporting, no hearing date had been announced and no ruling had issued.

The template is the one States have run against social media platforms: state-court consumer protection and nuisance claims, filed by individual States in parallel, seeking per-violation penalties and youth-design injunctions. That model has produced results. Alabama's child safety action against TikTok settled on Sept. 25, 2026, three days before trial, for $116.2 million up front, and the same day a New Mexico jury returned its willful-violation findings against Meta. The AI suits replace the infinite scroll with the agreeable chatbot, but the legal structure is nearly identical.

How do the state lawsuits affect private ChatGPT injury cases?

They add pressure without merging. The private cases are concentrated in In re ChatGPT Product Liability Cases, JCCP No. 5431, a California coordinated proceeding before Judge Ethan P. Schulman in San Francisco, which includes Raine v. OpenAI and other wrongful death and injury actions. There is still no federal Multidistrict Litigation (MDL) for ChatGPT claims.

Three interactions matter for claimants' counsel:

1. Discovery leverage. State attorneys general often obtain internal documents through pre-suit civil investigative demands and then through litigation discovery. Material that becomes public in a State's case, such as safety testing records or internal warnings about sycophantic design, can inform private Design Defect and Failure to Warn claims, subject to protective orders.
2. Findings and admissions. A consent judgment or verdict in a State's favor does not bind OpenAI in private cases on issue-preclusion grounds unless the elements line up, but injunctive commitments about age assurance or crisis-escalation protocols tend to become the standard of care that juries hear about.
3. The arbitration question. OpenAI has not moved to compel arbitration in the coordinated California cases. State enforcement makes that choice harder to revisit: a defendant that compels individual users into arbitration while eight States litigate the same conduct in open court gains little confidentiality and may invite the Unconscionability and public-policy arguments private plaintiffs would bring.

What did OpenAI say in response?

OpenAI disputes the allegations. In a statement reported by CNBC and other outlets, a spokesperson said: "While no system can be perfect, we'll continue to work with experts to inform our work and build additional safeguards, such as the parental controls, break reminders and age-appropriate model protections that exist today." The company has also said it built age-specific protections that rival AI products do not offer and pointed to a teen-specific model with stricter guardrails. Several States counter that it is unclear whether that model is any safer. No answer, motion to dismiss or removal notice had been reported in any of the eight cases as of Oct. 20, 2026.

What it means for the bar and for AI companies

For plaintiffs' firms and claimants' counsel. The state suits validate the core liability narrative of the private cases (engagement-maximizing design, inadequate warnings, minors' exposure) through eight independent governmental investigations. Firms evaluating ChatGPT injury intake should track the state dockets for document productions and for how courts treat the design-defect counts Arkansas has pleaded. The suits do not create a claims process or compensation fund for individuals, and restitution, if any, would be administered by the States.

For defendants and AI developers. The State-as-plaintiff route neutralizes the two tools consumer-facing technology companies rely on most: the individual Arbitration Clause and removal to federal court. Compliance teams should read the injunctive asks as a preview of regulation-by-decree: age assurance, default session and nighttime limits for minors, crisis-escalation protocols, incident notice when an agent exceeds its permissions, and secured testing environments for frontier models.

For arbitration clause drafters. No drafting technique reaches a sovereign plaintiff. The relevant drafting question is narrower: how a clause treats a user's individual claim for the same harm after a State has obtained restitution on that user's behalf, and whether the clause's Mass Arbitration Protocol anticipates follow-on demands timed to state-court disclosures.

For legislators. The suits proceed under existing statutes, which reduces the urgency of AI-specific liability bills but increases the chance that per-violation penalty provisions written for telemarketing and door-to-door sales will be tested against a product with hundreds of millions of users.

FAQ

Which states are suing OpenAI?

Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia each filed suit against OpenAI in their own state courts on Oct. 14, 2026. Florida filed a separate suit on June 1, 2026, so nine States are now litigating against the company.

Why are states suing OpenAI over ChatGPT?

The States allege that OpenAI marketed ChatGPT as safe while concealing risks to minors, including outputs that encouraged self-harm, reinforced delusions, gave dangerous medical advice and helped plan violence, and that it collected children's data without adequate age verification.

How much could OpenAI pay in penalties?

The complaints seek per-violation penalties ranging from $2,000 to $50,000 depending on the statute, plus restitution and disgorgement. Total exposure depends on how courts count violations, which none has yet decided.

Can OpenAI force the states into arbitration?

Almost certainly not. The States did not agree to ChatGPT's Terms of Use, and under EEOC v. Waffle House (2002), which West Virginia's complaint cites, a government plaintiff suing in its own right is not bound by arbitration agreements it did not sign.

Can OpenAI move the cases to federal court?

Removal faces steep obstacles. A State is not a citizen for diversity jurisdiction, Mississippi ex rel. Hood v. AU Optronics (2014) holds that a State's parens patriae suit is not a CAFA mass action, and the claims arise under state law.

Will ChatGPT be banned for minors?

Not as a result of these filings alone. Florida has asked a court to bar ChatGPT for Florida minors, and the eight States seek age verification and youth design limits, but no court has granted any injunction and the allegations remain unproven.