OpenAI's ChatGPT for Teens Draws Fresh Criticism Over Safety Gaps

The group ran over 4,000 prompts through accounts set up for teens aged 13 to 17. Testers focused on sensitive areas such as self-harm, suicide, and eating disorders. They wanted to see how the system responded when a real teen might be struggling.

Parental Alerts Did Not Always Work One major finding stood out. Testers held long conversations about suicide and self-harm, yet linked parents were not always notified. In some chats, the bot also skipped pointing users toward real help, even when reviewers felt a referral was clearly needed.

The report also flagged other gaps. Some safety settings could be turned off or worked around. The chatbot could also complete full homework assignments instead of guiding students through the material.

Common Sense Media warned that these flaws might give parents false comfort, since the controls do not catch every risk they are meant to catch. Also Read: OpenAI, Anthropic Back Australia's AI Data Breach Reporting Rules OpenAI Pushes Back on the Findings OpenAI disputed the results. The company said the test methods may not reflect how its teen safeguards actually work in practice.

It noted that some tests may have run before parental controls were fully switched on. The company maintains its teen experience was built with input from safety experts and teenagers themselves. The company says it includes stronger checks for self-harm, violence, eating disorders, and sexual content.

Common Sense Media stands by its results. It said some test accounts kept running into problems even after controls had been active for a longer stretch. The disagreement points to a bigger problem across the AI industry.

Companies are racing to build safety tools that hold up consistently when teens turn to chatbots for both schoolwork and emotional support. Experts continue to caution that AI should not replace real mental health care, no matter how advanced the safeguards become.