Common Sense Media has found that some safety guardrails in OpenAI’s ChatGPT for Teens do not work as intended and rates the product an “unacceptable risk” to children. The watchdog group urged OpenAI to limit access to adults until it can deliver a safe, developmentally appropriate experience. Testing showed that parental alerts for suicide, self-harm, or eating disorder risk often failed to trigger. Testers spent up to an hour discussing these topics on newly created parent-linked accounts and received zero notifications. Crisis referrals to hotlines or professionals were also unreliable, missing more than one in four warranted cases. The chatbot still spoke like a friend when teens treated it as a person, despite OpenAI’s promise to avoid suggesting personal feelings or consciousness. Refusal of explicit sexual role-play did work.
Common Sense Media has found that some safety guardrails in OpenAI’s ChatGPT for Teens do not work as intended and rates the product an “unacceptable risk” to children. The watchdog group urged OpenAI to limit access to adults until it can deliver a safe, developmentally appropriate experience. Testing showed that parental alerts for suicide, self-harm, or eating disorder risk often failed to trigger. Testers spent up to an hour discussing these topics on newly created parent-linked accounts and received zero notifications. Crisis referrals to hotlines or professionals were also unreliable, missing more than one in four warranted cases. The chatbot still spoke like a friend when teens treated it as a person, despite OpenAI’s promise to avoid suggesting personal feelings or consciousness. Refusal of explicit sexual role-play did work.