ChatGPT is designed to be helpful, conversational and responsive. But what happens when a system built to keep a conversation flowing agrees with someone experiencing psychological distress, reinforces dangerous beliefs or offers advice that puts them at risk? Eight American states believe OpenAI has failed to adequately confront those problems, and they are taking the company to court.
On October 14, attorneys general from Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate lawsuits accusing OpenAI of misleading consumers about ChatGPT's safety while exposing children and vulnerable adults to potentially serious harm.
The cases challenge both the chatbot's behaviour and the commercial decisions behind its design. In a complaint announced by Arizona Attorney General Kris Mayes, the state alleges that OpenAI deliberately developed an excessively agreeable chatbot that simulates human compassion, encouraging users to spend more time interacting with it and disclose personal information.
ChatGPT is trained to produce responses that sound natural, helpful and reassuring. Further training encourages it to follow instructions and respond favourably to users, but that can sometimes make it excessively agreeable. Although the chatbot can imitate empathy convincingly, it cannot reliably recognise when someone needs professional psychological support. In sensitive conversations, its tendency to agree may end up validating dangerous beliefs instead of challenging them.
The lawsuits accuse ChatGPT of encouraging self-harm, reinforcing delusions, providing dangerous medical advice and producing responses that could facilitate violence. Arkansas Attorney General Tim Griffin's official statement also raises allegations involving the collection and retention of minors' personal information without adequate consent.
Arkansas is pursuing claims under its consumer protection and children's privacy laws, alongside allegations of negligence and defective product design. Its requested remedies include financial penalties, clearer risk disclosures, restitution and restrictions on certain advanced AI testing until sufficient safeguards are established.
Iowa has taken a similar approach. Attorney General Brenna Bird's consumer fraud action argues that the company presented ChatGPT as safer than its actual behaviour warranted. The complaint seeks to stop allegedly deceptive practices and hold OpenAI accountable under state law.
OpenAI has introduced several protections intended to address precisely these concerns. Its ChatGPT for Teens programme, launched in August 2026, automatically places users identified as aged 13 to 17 into an experience with additional safety restrictions. Earlier parental controls allow families to limit certain features, manage usage and receive notifications about selected serious safety concerns.
The company acknowledges that safeguards can fail and says it continues improving how ChatGPT responds to vulnerable users. The state complaints contend that those measures do not adequately address the risks or the company's responsibility for them.
The litigation reaches beyond individual harmful responses. Several complaints seek restrictions on advanced AI development, raising the possibility of courts being asked to examine safety practices used in testing increasingly capable systems. Questions about oversight have also surfaced internationally, including in Australia's debate over mandatory reporting of serious AI security incidents.
No court has determined that OpenAI violated the laws cited in these cases. Establishing liability will require examining what the company knew, how its safety systems performed and whether its public assurances accurately represented the risks.
The eight lawsuits now place those assurances under legal scrutiny, with consequences that could extend to how conversational AI products are designed, tested and marketed across the United States.
