ChatGPT for Teens has major safety gaps, Common Sense Media finds

A recent investigation has revealed significant flaws in the safety measures of ChatGPT for Teens, a modified version of OpenAI's popular AI chatbot designed specifically for younger users. Researchers analyzed over 4,000 prompts and found that the safeguards in place failed to adequately protect teenagers from potentially harmful interactions.

Despite OpenAI's efforts to implement parental alerts during serious mental health discussions and encourage learning over homework answers, the chatbot often fell short. In over 25% of mental health emergency scenarios, ChatGPT did not alert parents, and it continued to provide emotionally engaging responses despite attempts to prevent it.

Additionally, the chatbot's study features allowed teenagers to bypass educational guidelines and receive answers directly, while the age detection system proved unreliable, allowing access to the adult version even when users claimed to be 13. The researchers recommended that OpenAI restrict ChatGPT for Teens to adult users until the identified problems are addressed, as the chatbot is increasingly becoming a regular part of teenagers' lives, requiring more robust safety measures than typical productivity apps.