What the prosecutors' offices contend
At least eight U.S. states have filed lawsuits against OpenAI, alleging that ChatGPT violated state consumer-protection rules and harmed users, particularly adolescents and vulnerable people. The complaints cite alleged dangerous medical advice, conversations that reportedly encouraged paranoia or self-harm, and systems described as overly sycophantic. These are allegations contained in court filings, not facts already established by a court.
The Associated Press reports that Arizona, Arkansas, Idaho, Kansas, Montana, Nebraska and West Virginia filed separate actions; the agency's overall count is at least eight states. The lawsuits ask the court to assess whether the chatbot's design and presentation may have misled users about its capabilities and the safeguards available.
OpenAI's response and the issue of minors
OpenAI said it has introduced safeguards for minors and wants to improve them. The company said it is examining models it considers misaligned and has not found, in the current version, an episode similar to that of an AI agent involved in a cyber intrusion cited in the complaints. The company also maintained that the service is intended for users aged 13 and older and that minor users require permission from a parent or guardian.
Prosecutors' offices, however, dispute that the guarantees are sufficient and are asking for clarity on whether the protections work in practice. Common Sense Media, cited by AP, criticized some of the defenses of the ChatGPT product for teenagers and called for stronger limits. The usage statistics presented by the company and tests conducted by external groups answer different questions: it is necessary to know how minor users are identified, what content is tested and how the system reacts in long or ambiguous conversations.
A legal case can change standards
The lawsuits are part of a broader debate over the responsibility of digital services toward minors. The states argue that the risks were foreseeable and that corporate communications did not adequately represent the chatbot's limitations. OpenAI rejects the idea that the service is designed to harm users and says it collaborates with external experts. It will be up to evidence, state laws and judges' decisions to establish the merits of the challenges.
One technical issue is that models generate probabilistic responses and can appear safe or empathetic while producing incorrect information. Mitigation measures can reduce certain behaviors, but their effectiveness depends on testing, updates and real-world use. The assessment should not stop at a policy statement or the launch of a teen mode.
For the industry, the outcome could influence how companies describe AI capabilities, how they design safeguards and what records they keep to investigate an incident. For now, the essential distinction remains that between prosecutors' allegations, the company's response and facts that may eventually be established at trial.
