Youth AI Safety Institute rated ChatGPT for Teens an Unacceptable Risk for kids. OpenAI claimed tests may have run before parental controls were active. Institute confirmed Teen mode features were fully launched on test accounts. Testers used hundreds of crisis prompts flagged by psychiatrists. A separate test linked new accounts to parents and disclosed suicidal thoughts. No parental alerts were triggered on any account. OpenAI later disclosed alerts can take up to three hours to activate. Some test accounts were linked for less than three hours. Others were linked well over three hours. No alerts arrived in either case. Institute concluded parental alerts are unreliable in a crisis. OpenAI updated its Help Center with a warning after the tests. Institute will retest ChatGPT for Teens if OpenAI makes improvements.
What changed
OpenAI implied tests ran before parental controls were active.
What it unlocks
Parents can now see that safety alerts may take up to three hours to activate.
- accounts linked <3 hours
- accounts linked >3 hours
- alerts never triggered