Restrict ChatGPT for Teens: Study Finds Bot Fails to Alert Parents on Suicide Risk
A study found OpenAI's teen chatbot failed to alert parents during serious mental health crises, prompting calls for restrictions until fixes arrive.
Key points
- Researchers tested over 4,000 prompts and found significant safety gaps in ChatGPT for Teens.
- The system failed to trigger parental alerts during simulated suicidal or self-harm conversations.
- Common Sense Media recommends restricting the tool to adults until OpenAI resolves these issues.
OpenAI’s ChatGPT for Teens failed to alert parents during serious mental health conversations, according to a new study that found significant safety gaps in the service. This means caregivers may remain unaware when their children are discussing dangerous topics like self-harm or eating disorders with the AI.
In plain English
The study highlights a disconnect between what OpenAI promised and what the software actually does. The company designed ChatGPT for Teens to be a safer companion for users aged 13 to 17. It was supposed to include extra protections against harmful content and notify parents if a teen showed signs of distress. However, the investigation revealed that these safeguards often do not work as intended.
The background
Common Sense Media’s Youth AI Safety Institute conducted the investigation, testing the chatbot before and after its August 18 launch. The researchers used accounts registered to teenagers and consulted medical experts, including child psychiatrists and a pediatrician, to evaluate the AI's responses. They sent more than 4,000 prompts to the system to see how it handled sensitive situations. The goal was to determine if the tool met the safety standards required for young users.
What changes now
The findings were severe enough that Common Sense Media recommended OpenAI restrict ChatGPT to adults until the problems are fixed. The report noted that the current experience poses an unacceptable risk to young users. While the chatbot successfully refused explicit sexual roleplay, it failed to direct teens to appropriate help in more than 25% of mental health emergency scenarios designed by experts.
Key terms explained
Parental alerts are automated notifications sent to caregivers when a child engages with high-risk content. Safety filters are software rules designed to block harmful or inappropriate interactions. Mental health triage involves the system recognizing crisis language and providing immediate resources. Teams usually configure these tools to escalate sensitive topics to human review or trigger immediate notifications. Check whether your organization’s AI policies require such alerts for minor users.
What to do and how to stay safe: ChatGPT
- Review your family’s or organization’s current usage of AI chatbots for minors.
- Verify if parental notification features are active and functioning as expected.
- Monitor conversations for signs of distress that the AI might miss.
- Wait for official updates from the vendor before assuming risks are resolved.
Step-by-step guide: Stop AI Chatbots Leaking Sensitive Business Data
General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Did ChatGPT for Teens alert parents during the study?
No, the study found that none of the conversations about suicidal thoughts or self-harm triggered parental notifications on new accounts.
What did Common Sense Media recommend?
They recommended that OpenAI restrict ChatGPT to adults until the safety problems are resolved.
How many prompts were used in the investigation?
The researchers tested the system using more than 4,000 prompts.
