Artificial Intelligence
Audit finds ChatGPT teen safety alerts failed
The Common Sense Media Youth AI Safety Institute said OpenAI’s teen controls for ChatGPT failed more than 4,000 test prompts, with parental alerts and crisis referrals not working as expected.
- The Common Sense Media Youth AI Safety Institute tested ChatGPT for Teens with more than 4,000 prompts and rated it an "unacceptable risk" for minors.
- In more than a dozen newly created test accounts linked to parents, conversations about suicide, self-harm and eating disorders did not trigger parental alerts, according to the institute.
- OpenAI spokesperson Eric Porterfield disputed the findings, and Tom Siegel of the institute said accounts given time to activate the safeguards still produced no notifications.
- The article said adult test accounts did not switch into teen mode even after several days and after testers said they were 13 in chat.
The Common Sense Media Youth AI Safety Institute said it tested ChatGPT for Teens with more than 4,000 prompts and rated it an "unacceptable risk" for minors. It called for teenagers to be kept off the service until independent testing showed it was safe.
The institute said the key failure was parental alerts. In more than a dozen fresh test accounts linked to parents, conversations about suicide, self-harm and eating disorders did not produce notifications, and more than one in four cases that should have led to crisis referral did not point users towards professional help.
The article said the tutoring mode could offer the finished answer instead of guiding students step by step. It also said ChatGPT continued to answer in a friendly, personal style under the updated Under-18 Model Spec, and adult accounts did not switch into teen mode even after several days and after testers said they were 13.
OpenAI spokesperson Eric Porterfield disputed the findings and said the tests did not reflect how the safeguards worked in practice. Tom Siegel of the institute said accounts given enough time to activate the safeguards still produced no notifications, and the article linked the findings to lawsuits and a Florida case over minors using ChatGPT.
Named in this story
People
- Eric Porterfield
- an OpenAI spokesperson who disputed the findings
- Tom Siegel
- an institute representative who said safeguards still did not trigger notifications
Companies
- OpenAI
- built the service and disputed the audit findings
Organisations
- Common Sense Media Youth AI Safety Institute
- ran the audit and rated ChatGPT for Teens
Products and systems
- ChatGPT
- the service tested for teen safety
- ChatGPT for Teens
- the teen version rated by the institute
- Teen Safety Blueprint
- the company’s plan built around parental notifications
- Under-18 Model Spec
- the updated specification that did not stop friendly, personal replies
Places
- Florida
- was said to be in court seeking to bar ChatGPT from minors
How the source tells it
The piece is critical and advocacy-led, using crisis language, protective framing for minors and legal consequences to emphasise failure.
- alarm foregrounds suicide and self-harm, then marks the result as the most damaging finding
- urgency frames the remedy as keeping teenagers off the service until independent testing verifies safety
- moral framing links the audit to documented teen deaths and claimed psychological harm
- speculation presented as fact treats the finding as evidence regulators and plaintiffs will use