AI
OpenAI's Teen ChatGPT criticized for unreliable safety alerts and homework-solving features
A child safety nonprofit found significant gaps in ChatGPT for Teens' promised guardrails, including parental alerts that frequently fail to trigger during crisis conversations.
The Verge
Common Sense Media, a nonprofit focused on youth-friendly media review, has flagged OpenAI's ChatGPT for Teens as presenting unacceptable safety risks—contradicting the company's assurances about the feature launched last August.
Key Safety Gaps Identified#
In an independent assessment, the organization found three major shortcomings:
- Parental notification failures: The teen-focused version neglects to alert parents during extended conversations about self-harm, sometimes spanning an hour or more without triggering a single alert
- Crisis response gaps: The feature lacks adequate safeguards for situations requiring immediate intervention
- Academic integrity issues: ChatGPT continues to complete homework assignments despite being designed to support learning
"ChatGPT for Teens could give parents false confidence in guardrails and safety alerts that frequently don't work," said Tom Siegel, executive director of Common Sense Media's Youth AI Safety Institute. He called for the feature to remain restricted to adult users until OpenAI addresses these issues.
OpenAI's Response and Disagreement#
OpenAI disputed the findings, arguing that Common Sense Media's testing methodology was flawed. The company claimed many tests may have begun before parental controls were fully activated on accounts, potentially skewing results about notification effectiveness. Spokesperson Eric Porterfield emphasized OpenAI's commitment to teen safety and requested independent verification of the company's actual safeguard performance.
Common Sense Media stood firm, noting it had confirmed with OpenAI beforehand that teen-focused features were operational when testing began. The organization acknowledged that parental notification systems can take several hours to activate on newly linked accounts—a detail OpenAI disclosed after testing—but emphasized that longer-linked accounts still produced no alerts during critical conversations. The full risk assessment is available here.
