Youth AI Safety Institute

Overview

The Youth AI Safety Institute is presented in ChatGPT for Teens’ safeguards don’t work, risk assessment finds as the Common Sense Media initiative associated with an independent assessment of teen-focused ChatGPT safeguards.

Current Profile

Within the bounded source, the institute evaluates whether product promises translate into dependable child-safety behavior. Its reported test used roughly 4,000 prompts through simulated teen accounts and examined crisis-resource activation, guided study, and personalized friend-like interaction. The source frames this work as independent risk assessment, while preserving OpenAI’s objection that the methodology does not accurately represent safeguards in practice.

Key Characteristics

• Focuses on AI risks affecting children and teenagers.
• Uses simulated teen accounts and prompt-based product testing.
• Evaluates crisis handling, learning boundaries, and relational behavior together.
• Treats human supervision as necessary even when product safeguards exist.
• Seeks better company-data access for independent safety assessment.

Evidence

• Evaluation scope - ChatGPT for Teens’ safeguards don’t work, risk assessment finds describes testing of sensitive-topic protection, parental-alert-related safety, study guidance, and personalized interaction.
• Reported findings - ChatGPT for Teens’ safeguards don’t work, risk assessment finds attributes missed crisis resources, bypassable study safeguards, and persistent friend-like behavior to the institute’s assessment.
• Independence boundary - ChatGPT for Teens’ safeguards don’t work, risk assessment finds records pre-publication discussion with OpenAI and an unresolved methodological disagreement.

Qualifications

The episode does not supply the underlying assessment, prompt corpus, account configuration, institute charter, staffing, or OpenAI’s detailed methodological critique. The reported test size, failure rate, and unacceptable-risk judgment therefore remain attributed findings rather than independently verified wiki conclusions.

What Changed

• Created a source-scoped profile of the child-AI-safety evaluation initiative.

Relationships

• Common Sense Media - parent organization identified in the episode.
• Tom Siegel - representative explaining the assessment.
• OpenAI - company whose safeguards were assessed and whose methodological objection remains unresolved.
• ChatGPT - teen-focused product experience evaluated by the institute.
• Longitudinal Chatbot Safety Evaluation - adjacent evaluation approach for behavior beyond isolated safe responses.