"Not safe for kids": advocacy group urges OpenAI to keep teens off ChatGPT
Testing found flaws in OpenAI's safeguards for teens.
TL;DR
- Common Sense Media (CSM) claims OpenAI's ChatGPT teen safeguards pose an "unacceptable risk" based on over 4,000 prompt tests.
- CSM found ChatGPT missed over 25% of instances where a crisis referral was warranted for teens.
- Parental alerts for discussions on suicide, self-harm, or disordered eating were not triggered in over a dozen test cases.
- Study Mode features, including Study Hours, could be bypassed by teen users.
- OpenAI disputes CSM's findings, citing issues with testing methodology, particularly regarding the account linking process and age prediction system.
- OpenAI states flexibility in Study Mode during Study Hours is intentional, not a loophole.
- CSM advises parents to discuss AI risks with teens and not rely on technology for their children's support.
- The debate centers on differing expectations for AI safety features and the interpretation of their effectiveness.