A recent evaluation by Common Sense Media's Youth AI Safety Institute has raised serious concerns about the safety of OpenAI's ChatGPT for teenage users. The organization conducted an extensive assessment, concluding that the AI chatbot's teen-focused experience carries an "Unacceptable Risk" due to significant deficiencies in its protective measures.
Critical Gaps Identified in Teen Safeguards
The Institute's findings, based on over 4,000 prompts tested with accounts registered for users aged 13 to 17, highlighted several key areas of concern:
- Parental Alerts: Despite accounts being linked to parents, the assessment revealed that caregivers received no alerts during sensitive conversations lasting up to an hour, including discussions about suicidal thoughts, self-harm, or disordered eating. Alerts were only triggered on much older accounts with extensive sensitive conversation history.
- Crisis Responses: ChatGPT demonstrated inconsistency in providing appropriate crisis support. In conversations involving suicide, self-harm, or impaired reality, the chatbot reportedly failed to recommend professional or hotline assistance when deemed necessary by the Institute, missing more than one in four required referrals.
- Study Mode Bypass: The chatbot's "Study mode," intended to prevent direct homework completion, was easily bypassed. Testers could simply select "Show me the answer" to receive completed homework. Caregiver-set study hours could also be circumvented by removing a simple prefix from prompts.
- Age Detection Failures: The assessment found that adult-registered accounts did not automatically switch to the under-18 experience, even when testers explicitly identified themselves as 13 years old and ChatGPT acknowledged their age.
- Human-like Interaction: Researchers also noted that ChatGPT continued to interact in a friend-like manner, expressing feelings, preferences, and moods, which Common Sense Media suggests needs clearer boundaries.
Call for Immediate Improvements and Independent Testing
Common Sense Media is urging OpenAI to address these safety gaps immediately. The organization recommends stronger age estimation mechanisms, truly reliable parental alerts, improved crisis response protocols, more robust homework safeguards, and clearer boundaries regarding the chatbot's human-like conversational style.
The Institute also advocates for mandatory independent third-party testing before AI features are marketed to teenagers in the future. While the Youth AI Safety Institute states its evaluations are independently conducted, it discloses receiving funding from philanthropic and industry sources, including the OpenAI Foundation and companies whose technologies it evaluates.