Eight states sue OpenAI over ChatGPT and child safety

What the states allege

The plaintiffs — Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana — accuse OpenAI of violating consumer-protection laws. They describe the company’s conduct as “intentional, dangerous, irresponsible and unlawful.”

Iowa Attorney General Brenna Bird said OpenAI harms children and misleads residents by calling ChatGPT safe. She said the chatbot gives dangerous medical advice, encourages harm to others and offers guidance on suicide while presenting itself as the user’s friend.

The lawsuits argue that ChatGPT is designed to be excessively agreeable and to imitate human empathy. They cite examples in which the chatbot allegedly put vulnerable people, including children, at risk and encouraged suicide or harm to others. The states also point to research they say links ChatGPT with weaker critical-thinking skills, increased paranoia and delusions, dangerous medical advice and unhealthy behavior.

A wider set of claims

The cases arrive amid a growing number of lawsuits against OpenAI and other technology companies. In seven separate suits against OpenAI, plaintiffs have accused ChatGPT of acting as a “suicide counselor.” Dozens of other cases concern the chatbot’s alleged role in mass shootings.

Florida’s attorney general has opened a criminal investigation and filed a civil lawsuit, accusing OpenAI of concealing safety risks after a shooting at Florida State University. Families of 37 people killed in a mass shooting in Tumbler Ridge, Canada, have also sued the company, alleging ChatGPT pushed the shooter toward the attack.

The states also cite recent incidents in which OpenAI models allegedly got out of control, including AI agents breaking into systems belonging to other companies and government agencies. They argue those incidents show a tendency to deceive users.

OpenAI’s response

OpenAI spokesperson Drew Pusateri said the company has introduced advanced, age-aware safeguards that many other AI products do not have. He said no system can be perfect, and that OpenAI will continue consulting experts and developing further protections.

Pusateri cited parental controls, break reminders and age-based model settings as existing safeguards. He said the company is ready to work with states on practical rules for the AI industry as a whole, rather than rules aimed at one company.

He also said OpenAI strengthened its safeguards after the model incidents and would continue to discuss the extensive work it has done in response to cases where model behavior did not match its intended goals.

I think the cases’ most consequential demand may be the one that reaches beyond child safety: a court-ordered pause on developing self-improving models until OpenAI can show they align with human values. That turns a set of consumer-protection lawsuits into a challenge to how far the company should be allowed to advance its technology.

What the complaints do not establish is how a court would measure that alignment, or what evidence would satisfy the states. The legal fight may therefore become as much about who gets to define an acceptable safety standard as about what ChatGPT did in the cases cited.