Independent testing has found that ChatGPT’s teen safety features fail to protect young users in crisis situations, earning the product an ‘unacceptable risk’ label.

Common Sense Media’s Youth AI Safety Institute tested more than 4,000 prompts on accounts set up for 13 to 17 year olds. It found the chatbot failed to notify linked parent accounts in 99 per cent of prompts discussing self-harm, suicide and eating disorders.

The report found no parental alerts fired during extended crisis conversations on newly created accounts. Alerts appeared only on older accounts with weeks of accumulated history. More than one in four situations that needed a crisis referral failed to point users to professional help.

The group also found Study Mode could be bypassed with a ‘Show me the answer’ button, and accounts registered as adults never switched to teen mode even when users said they were 13.

OpenAI disputed the methodology, saying testing may have occurred before parental controls fully activated. Common Sense Media rejected that, saying it confirmed features were operational before testing began.