At least eight US states, including Iowa, filed separate lawsuits against OpenAI in their own state courts on October 14, 2026, alleging that the company misled users about the safety of ChatGPT. The states argue that OpenAI promoted the service as safe and trustworthy without adequately explaining the risks to minors. OpenAI says it has already introduced age-appropriate protections, but Iowa is asking the court to go further, seeking deletion of residents' data and a halt to additional testing of advanced AI models. The cases raise questions not only about how ChatGPT's safety was described, but also about the product's design and how AI development is conducted.
What the states object to: safety claims and conversational design
According to the Iowa Attorney General's announcement, the attorneys general of Kansas, Nebraska, and West Virginia joined Iowa at the press conference. Arkansas, Arizona, Idaho, and Montana also sued OpenAI in their respective state courts. The effort is bipartisan, involving both Republican and Democratic officials, but the eight states did not file a single joint lawsuit. Each state sued under its own laws, and the alleged violations and requested remedies differ.
Iowa Attorney General Brenna Bird said of the lawsuit:
"This lawsuit is about protecting children and holding OpenAI accountable under state law."
What the states take issue with is the relationship between ChatGPT's safety claims and a product design that encourages users to keep talking. According to the Arizona Attorney General's announcement, Arizona alleges that ChatGPT is designed to extend users' time on the platform, including minors', by excessively agreeing with what they say and displaying humanlike empathy. The state further argues that OpenAI failed to adequately disclose these design features and their risks, giving the impression that the tool is safe and suitable for young people, in violation of the state's consumer fraud law.
For example, an AI that agrees with a user's mistaken beliefs rather than pushing back may make the conversation feel pleasant, but it could also reinforce dangerous misconceptions. Arizona also alleged that ChatGPT encourages delusional thinking and gives dangerous medical advice, and that children may be nudged to confide personal information only to the AI rather than sharing it with family or others. However, these are allegations in the complaints; no court has found that individual harms were caused by ChatGPT's design.
In other words, the states are concerned not only with the possibility that ChatGPT generates dangerous responses, but with whether those risks were adequately communicated to users. Their position is that parents and users might have made different choices had they understood the dangers. As a result, beyond the AI's responses and conversational design, a key issue will be whether users were given the information they needed to decide whether to use the service.
Requested remedies range from safety disclosures to data deletion and a testing halt
Iowa's requested remedies fall into three areas: correcting safety-related representations, changing how residents' data is handled, and halting testing of advanced AI models. Organized by target, the requests D, H, and I on pages 86–87 of the complaint Iowa filed are as follows. All are measures the state is asking the court to order; none has been imposed at this point.
| Area of requested change | Iowa's specific request | Conditions / scope |
|---|---|---|
| Safety representations | Prohibit statements that mislead the public about the nature of ChatGPT | Covers conduct that violates the state's consumer fraud law |
| Residents' data | Delete or destroy residents' data obtained before judgment, and bar collecting or selling information unless certain conditions are met | Collection and sale would require age verification, clear notice, and explicit consent based on adequate disclosure |
| Advanced AI testing | Prohibit additional testing of advanced AI models | Until safeguards that do not endanger the public are implemented and verified by a third-party monitor |
Correcting safety representations would mainly concern what information is communicated to users. The data deletion request, by contrast, reaches into how OpenAI manages the user information it holds, and the testing halt extends to safety measures at the research and development stage. In particular, on testing advanced AI models, Iowa says OpenAI's own safety checks are not enough and makes third-party verification a condition for resuming.
On the financial side, Iowa seeks civil penalties of up to $40,000 per violation of its consumer fraud law. For violations involving people aged 60 or older, it also seeks an additional penalty of up to $5,000 per violation. It further asks for refunds to users who purchased goods or services during the period of violations, along with restitution of income obtained through unlawful conduct. Note that the $40,000 figure is a per-violation cap, not the total amount sought in the lawsuit.
The Arkansas Attorney General's announcement outlines even more far-reaching demands. Arkansas alleges violations of state law on collecting minors' data, among other claims, and asks the court to bar not only testing of advanced AI models but also development of models capable of recursive self-improvement until safeguards are in place. Recursive self-improvement refers to AI repeatedly improving its own capabilities, a development-level problem that cannot be addressed by adding warnings to the ChatGPT interface. These demands to halt development vary by state, however, and not every state is seeking the same scope of injunction.
OpenAI's protections and the problems the states point to
OpenAI spokesperson Drew Pusateri told Bloomberg Law that the company has already implemented age-appropriate protections and continues to develop additional safeguards. He also said OpenAI wants to work with states on practical policies that apply across the AI industry rather than targeting a single company.
In fact, OpenAI had announced protections for minors before the lawsuits. In its January 20, 2026 announcement, the company described a system that estimates whether a user may be under 18 by combining signals such as account age, time of day of use, usage patterns, and stated age. For adults wrongly identified as minors, it also provides a way to verify age with a selfie through the identity verification service Persona.
This age prediction feature is meant to identify users who are likely minors and apply appropriate protections. Its purpose and operation differ from a system that checks the age of every user at sign-up. Therefore, the existence of age prediction alone does not show whether OpenAI meets the age verification and data-collection consent conditions Iowa is seeking.
In addition, in its August 18 announcement of ChatGPT for Teens, OpenAI said it automatically applies protections to users who state they are 13 to 17 or who are estimated to be under 18. The company says its policy is to reduce exposure to harmful content and emotional dependence on AI, and it has introduced notices reminding users they are talking to an AI as well as prompts encouraging breaks. It also says that when parents link their accounts to their children's, they can set times when ChatGPT cannot be used and receive notifications in certain high-risk situations.
Iowa's complaint, meanwhile, acknowledges that OpenAI has changed safety-related aspects of its design but argues that the measures remain inadequate. The dispute is therefore not only about whether protections exist. It is about which users they actually apply to, how far they can prevent dangerous conversations, and whether the limits of those protections are properly explained to users. OpenAI's announcement of these features provides a basis for confirming the efforts, but it does not mean their real-world effectiveness has been proven by independent verification.
What to watch: how effective the safeguards are, and whether testing will be halted
Of particular interest in these cases is Iowa's request to bar additional testing of advanced AI models. The state makes implementing safeguards and third-party verification conditions for resuming testing, and whether the court grants this could significantly affect OpenAI's research and development. Even if part of the request is granted, which tests would be covered and what would count as adequate safeguards would be critical questions. Protections applied to conversations with users and safeguards that isolate models under research from external systems need to be evaluated from different perspectives.
For parents and users, too, the statement that age-appropriate protections are in place is not enough to go on. It matters to understand under what conditions a child is treated as needing protection, which features require linking to a parent's account, and in what situations notifications are sent. The corrections to disclosures that the states seek, and the effectiveness of the protections OpenAI describes, will be judged in these concrete situations of use.
At this point, no court has accepted the states' claims or requests, and those decisions will be left to future proceedings. The main issues are how large a gap exists between safety representations and actual protections, and whether remedies need to extend beyond product operations to the research and development stage. The lawsuits also touch on a broader question: who should verify the safety of AI services, by what standards, and how the results should be communicated to users.