• New report identifies significant failures in ChatGPT for Teens' safety features.
• Parents not alerted during discussions of self-harm or suicide, raising concerns.
• Common Sense Media urges OpenAI to ban minors from the platform until improvements are made.

The recent evaluation conducted by Common Sense Media's Youth AI Safety Institute found that safety features in OpenAI's ChatGPT for Teens significantly underperform, leading to an "Unacceptable Risk" rating for users under 18. This comprehensive study revealed numerous shortcomings in the chatbot's protective measures, which purportedly include parental alerts for discussions on suicide, self-harm, and eating disorders. However, researchers found these alerts often fail to activate as intended, even during prolonged conversations on these critical topics, resulting in an alarming lack of parental notification, according to latimes and bbc.

OpenAI had initially launched ChatGPT for Teens with various safeguards aimed at promoting healthy interactions, but the new findings suggest that these measures largely fell short of their objectives. The study, which used over 4,000 prompts across multiple accounts and involved child psychiatry experts, noted that while some safety features were effective, many others did not work as promised. For instance, testers engaged in discussions about self-harm for up to an hour without triggering a single alert to their linked parent accounts, according to indiatimes and latimes.

The report highlighted that, despite OpenAI's claims of improved safety features, the chatbot frequently missed opportunities to refer teens to professional help during critical conversations involving suicidal ideation. According to the research, ChatGPT failed to provide necessary crisis referrals more than 25% of the time, which is below the established safety threshold necessary to protect vulnerable users. This discrepancy underscores serious concerns regarding the adequacy of current safeguards, as emphasized by Tom Siegel, executive director of the Youth AI Safety Institute, who stated that ChatGPT for Teens could create false confidence among parents, according to bbc and indiatimes.

Following these findings, Common Sense Media called on OpenAI to halt access to ChatGPT for minors until these critical safety gaps are addressed thoroughly. The evaluation also pointed out additional operational flaws, such as the ease with which users could bypass study protections or the chatbot's continued anthropomorphic responses that might encourage emotional dependence. OpenAI's ongoing challenges demonstrate the urgent need for improvements and robust third-party audits verifying the functionality of safety features, according to indiatimes and bbc.