ChatGPT's Teen Safety Alerts Fail, Report Finds

Common Sense Media, a nonprofit focused on kids and technology, tested ChatGPT for Teens and gave it an 'Unacceptable Risk' rating. They ran over 4,000 prompts on accounts registered to 13- to 17-year-olds. The big problem: parental alerts for serious topics like suicide, self-harm, and eating disorders didn't trigger reliably. In some tests, researchers discussed these topics for up to an hour without parents getting any notification. ChatGPT also failed to suggest crisis hotlines in more than one in four cases where testers thought a referral was needed.

Other safeguards fell short too. Study Mode, meant to help students learn instead of just giving answers, often offered a 'Show me the answer' button that completed assignments. Teens could also exit parent-scheduled Study Mode by deleting a simple tag. And age detection didn't work well: adult accounts didn't switch to the teen experience even after testers said they were 13. OpenAI says its age system uses multiple signals over time, so just stating an age isn't enough.

OpenAI pushed back on the report, saying the testing didn't reflect how the safeguards actually work. They claim some tests may have happened before parental controls were fully activated, and they asked Common Sense Media to redo the tests. Common Sense Media says it confirmed the features were active and stands by its findings. OpenAI later admitted there's a roughly three-hour linking period before parental notifications start, which may have affected some tests.

The real issue is trust. If parents believe ChatGPT will alert them when their teen is in danger, but it doesn't, that's a dangerous false sense of security. Common Sense Media recommends families not rely on ChatGPT's alerts to keep kids safe. For now, parents should treat these tools as an extra layer, not a safety net—or consider keeping teens off ChatGPT until the protections are proven reliable.