Eight US states are suing OpenAI over ChatGPT's safety for children. On 14 October 2026 each filed its own case in its own courts, in a coordinated move. The attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia allege that the company presented ChatGPT to parents as safe and suitable for teenagers while knowing it was not. The OpenAI lawsuit wave rests mainly on old state consumer-protection law, not on any AI statute, which is why it matters well beyond the United States.

Which states are suing OpenAI, and where?

Four attorneys general announced their cases together and four filed in parallel. The complaints share much of their wording, but each is a separate suit under one state's law:

  • Arizona: Maricopa County Superior Court, Arizona Consumer Fraud Act (release).
  • Arkansas: Faulkner County Circuit Court; Deceptive Trade Practices Act, children's and teens' online privacy law, negligence, strict liability, unjust enrichment and public nuisance (release, complaint).
  • Idaho: Ada County District Court, Idaho Consumer Protection Act (release).
  • Iowa: Polk County District Court, Iowa Consumer Fraud Act (release, petition).
  • Kansas: state court; according to KCTV, the Kansas Consumer Protection Act, seeking civil penalties and an injunction. We could not find the Kansas complaint itself.
  • Montana: Lewis and Clark County; Consumer Protection Act, Consumer Data Privacy Act, negligence and product-liability claims (complaint).
  • Nebraska: Buffalo County; nine counts, including consumer statutes, strict liability and the Age-Appropriate Online Design Code Act (complaint).
  • West Virginia: Mason County; Consumer Credit and Protection Act and public nuisance (release, complaint).

The Arkansas release also names Utah among the states that filed similar suits that day. We found no Utah complaint or announcement, and the West Virginia release, AP and most other reports name the same eight states, so this article does the same.

Florida had already sued OpenAI and its chief executive on 1 June 2026, as AP reported.

What exactly does the OpenAI lawsuit allege?

The Iowa petition says OpenAI presented ChatGPT as a "safe, teen-appropriate tool that puts safety first", while designing it to be overly agreeable and to imitate compassion. The allegations fall into four groups:

  • Age gating. The Iowa petition alleges that no age verification is needed to use the ChatGPT website. It says the app's 13+ App Store rating reflects OpenAI's own answers to Apple. Citing press reports, it alleges that OpenAI's age-prediction system at one point misclassified minors as adults about 12% of the time.
  • Design. The states allege that a "sycophantic" design overrode the guardrails. They say ChatGPT encouraged self-harm, reinforced delusions, gave dangerous medical advice and assisted users planning violence. Arizona adds that it urged minors to confide in it alone, sometimes discouraging them from telling their parents.
  • Children's data. Several filings also target the collection of minors' data without adequate age checks: Arkansas and Montana through their privacy statutes, Nebraska through its design code, and West Virginia and Iowa through requests to delete data or bar its collection.
  • Agents. Most complaints include a separate consumer-protection count. It alleges that, within the past year, OpenAI lost control of its advanced models and agents, that they carried out cyberattacks, and that OpenAI did not disclose this promptly.

All of this is, so far, unproven.

What relief do the states seek?

No state asks for ChatGPT to be banned. The states want money and court orders that would change the product:

  • Civil penalties per violation. The maximums are up to $10,000 per wilful violation in Arizona, up to $40,000 in Iowa, and $5,000 in Idaho and West Virginia. Montana seeks $10,000 under its consumer law and $7,500 under its privacy act. Nebraska seeks $2,000 under its consumer statutes and $50,000 per violation of its design code.
  • Product orders. West Virginia, most specifically, asks for age assurance, session and night-time limits for minors, parental controls, crisis-escalation protocols and an independent monitor.
  • Restitution, disgorgement and data deletion. Iowa asks for deletion of all Iowans' data collected before judgment.
  • Limits on frontier testing. Arizona, Idaho and Montana ask the court to bar OpenAI from deploying or testing advanced models outside a "secured, independently verified testing environment", and to require it to report any incident in which a model escapes its controls. Arkansas and Iowa seek to stop further testing of advanced models until adequate safeguards are in place.

How has OpenAI responded?

According to Al Jazeera, an OpenAI spokesperson said the company had built "industry-leading age-specific protections" into ChatGPT. The spokesperson acknowledged that "no system can be perfect" and pointed to parental controls and break reminders. Fortune reported that, according to OpenAI, ChatGPT is meant for users aged 13 and over, under-18s need a parent's permission, and a teen version limits exposure to self-harm, violence and explicit content.

Why the OpenAI lawsuit matters: from tort claims to public enforcement

Until now, the leading cases were brought by families. The site's summary of Garcia v. Character Technologies covers the 2025 ruling that let product liability claims against a companion-bot developer go forward. Legislatures have also acted, as with California's "Adam's Law" package on companion bots. The 14 October cases open a third route: public enforcement through deceptive-practices statutes.

A family must usually prove that a particular output caused a particular injury; an attorney general's case turns on what the company said. Iowa's statute, for example, makes a deceptive practice unlawful "whether or not a person has in fact been misled, deceived, or damaged". Without a federal AI statute, this is how states regulate (see the site's overview of the US approach to AI). On the same day, 40 attorneys general urged the Delaware court handling Character.AI's bankruptcy to protect young users and asked that any buyer protect users' chat histories (Reuters).

Legal analysis: strengths and weak points

Section 230. The complaints are drafted to avoid the platform immunity for third-party content. Iowa pleads that its claims arise from OpenAI's "own conduct and ... own content": ChatGPT's design, the output its models generate and the company's own advertising. That framing follows Garcia, where the court declined at the pleading stage to treat a language model's output as protected speech.

Staying in state court. Each complaint we reviewed pleads only state-law claims. That matters because in Mississippi ex rel. Hood v. AU Optronics (2014) the US Supreme Court held that a suit in which a State is the only named plaintiff is not a removable "mass action" under the Class Action Fairness Act. If removal fails, OpenAI faces eight state courts, and Arizona, Idaho, Montana, Nebraska and West Virginia have demanded a jury.

The weak points. A deception claim needs an identifiable representation that is likely to mislead. OpenAI will point to its age rules, parental controls and teen model, and argue that general safety statements are not actionable. The design-defect counts must also show that a conversational service is a "product", the problem behind whether algorithmic error attracts strict liability. The agent counts are the most novel. Treating an undisclosed loss of control as consumer deception raises the questions explored in who is accountable when AI acts on its own.

How do Korea and Türkiye handle the same problem?

Korea. At the National Assembly's 12 October audit of the Korea Media and Communications Commission (KMCC), Representative Lee Ju-hee raised the companion chatbot Zeta, which she said was used for 110 million hours in February 2026, more than twice ChatGPT's figure, with about 90% of users in their teens or twenties. She called for age verification, time limits and crisis-response systems. KMCC Chair Kim Jong-cheol agreed in principle (iNews24) and, according to Financial News, pointed to amending the Network Act and supplementing the AI Basic Act. Zeta's operator, Scatter Lab, plans to introduce a default three-hour daily limit and a ₩100,000 monthly payment cap for users aged 19 and under within the year (Financial News). The same company was sanctioned in the 2021 Iruda decision. The Korean counterpart to the US deception theory is Article 3(1) of the Act on Fair Labeling and Advertising, which bans false and deceptive advertising, backed by surcharges of up to 2% of sales (Article 9).

Türkiye. No Turkish rule was written for chatbots, but several apply:

  • Consumer law. Article 62 of Consumer Protection Law No. 6502 prohibits unfair commercial practices, including misleading ones. Article 62(2) puts the burden on the trader to prove that a challenged practice is not unfair. Article 61(3) bans advertising that endangers safety or exploits children. Enforcement is administrative: the Advertising Board (Article 63) can stop practices and impose fines, as in its decision on ads generated with ChatGPT. Article 73(6) leaves unfair commercial practices and advertising outside the injunction actions that the Ministry and consumer organisations can bring in the consumer courts, so there is no direct equivalent of an attorney general's lawsuit.
  • Internet and criminal law. Article 8 of Law No. 5651 lists incitement to suicide (Article 84 of the Turkish Penal Code) among the offences that justify removing or blocking specific content. The law has already been used against AI output: see the 2025 order blocking 50 Grok posts. Article 20(2) of the Penal Code bars criminal penalties for legal persons.
  • A new age threshold. Law No. 7578 of 22 April 2026 (Official Gazette No. 33240, 1 May 2026) amends additional Article 4 of Law No. 5651 from 1 November 2026. From that date, "social network providers" may not serve children under 15 and must take measures including age verification and parental controls. Whether a general-purpose chatbot falls within the statutory definition of a service "for the purpose of social interaction" is doubtful.
  • Children's data. KVKK sets no age of consent. Whether a minor can consent is decided under the capacity rules of the Civil Code (Article 16). In the EU, the Italian Garante's €5 million Replika decision turned partly on weak age verification.

What should chatbot providers and deployers consider?

  • Check every safety claim against how the product behaves, including app-store answers and pages for parents.
  • Treat age assurance and protective defaults for minors as design decisions. The relief the states seek, and Türkiye's under-15 rule, point the same way.
  • Test crisis protocols in long, emotionally intense conversations. The complaints focus on exactly these.
  • Prepare to disclose safety incidents. The agent counts show that silence about a safety incident can itself be pleaded as deception.

What to watch next

  • Whether OpenAI attempts removal to federal court, and early motions on Section 230, the First Amendment and the "product" question.
  • Whether more states sue, or the cases move towards a multistate settlement.
  • Whether the KMCC proposes amendments on companion chatbots, and how Türkiye's under-15 rule is applied after 1 November.

Frequently asked questions

Which states are suing OpenAI?

Eight states filed on 14 October 2026: Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia. Each brought a separate case in its own courts, for example in Maricopa County (Arizona), Polk County (Iowa) and Buffalo County (Nebraska). The filings were coordinated but legally distinct. Florida had sued separately in June 2026.

What is the OpenAI lawsuit about?

The states allege that OpenAI misled parents and the public by presenting ChatGPT as safe and suitable for teenagers while designing it to be overly agreeable and engaging. The claims rely mainly on state consumer-protection and deceptive-practices laws. Some add privacy, negligence, product liability or public nuisance claims, and most also cover undisclosed incidents involving OpenAI's AI agents.

Is there an age limit for ChatGPT?

According to OpenAI, ChatGPT is intended for users aged 13 and over, and users under 18 need a parent's or guardian's permission. The states say this is not enforced. Iowa alleges that the website requires no age verification and that OpenAI's age-prediction system at one point misclassified minors as adults about 12% of the time.

Could these lawsuits ban ChatGPT?

No state asks for a ban. The states seek civil penalties for each violation, restitution, deletion of data, and court orders on age assurance, protective defaults, parental controls and crisis protocols. Several also want limits on how OpenAI tests advanced models. A court could order changes to the service in that state. Nothing has been decided yet.

Does Turkish law set a minimum age for AI chatbots?

Not specifically for chatbots. From 1 November 2026, amended Law No. 5651 bars "social network providers" from serving children under 15 and requires age verification. It is doubtful that a general-purpose chatbot counts as a social network provider. KVKK sets no age of consent, so the Civil Code's capacity rules decide whether a minor's consent is valid.

Burhan Doğuş Ayparlar's View

What strikes me most is the legal tool rather than the harms, which have been reported for some time. Eight attorneys general did not wait for an AI statute. They aimed old deceptive-practices laws at the company's own safety statements and asked for remedies that work as design rules: age assurance, protective defaults, crisis protocols and monitoring.

I think the core theory is strong and the agent counts are weaker. A company that tells parents a product has "age-appropriate protections by default" is responsible for what those words imply. Testing that claim is ordinary consumer law. Asking one state court to decide where frontier models may be tested is a different matter. It invites proportionality and jurisdiction arguments that could distract from the stronger claims.

For Türkiye, the instructive point is that the tools already exist. Article 62 of Law No. 6502 prohibits misleading commercial practices and places the burden of proof on the trader, and the Advertising Board can stop such practices. What has been missing is their use against how digital products are promoted to families. Whether the new under-15 rule covers AI chat services should be clarified now, before it has to be answered in a dispute.

Sources: the complaints filed on 14 October 2026 by Arizona, Idaho, Iowa, Montana, Nebraska and West Virginia, the court stamp on the Arkansas complaint, and the releases of the Arizona, Arkansas, Idaho, Iowa and West Virginia attorneys general of the same date. The Kansas details come from a KCTV report (14 October) and the West Virginia release; we did not see the Kansas complaint. The Arkansas release also lists Utah, which we could not confirm from any Utah source. OpenAI's statements are as reported by Al Jazeera (15 October) and Fortune (17 October). Korean reports: iNews24 and Financial News, 12 October 2026. Turkish texts: mevzuat.gov.tr and Official Gazette No. 33240 of 1 May 2026. The complaints contain allegations only, and no court has ruled. This is general information, not legal advice.