CR4T: AI Safety Framework for Adolescent LLM Interactions

A new research paper from arXiv proposes CR4T (Critique-and-Revise-for-Teenagers), a model-agnostic safeguarding framework for large language models (LLMs) used by adolescents. The authors argue that current safety mechanisms, based on adult-centric norms and refusal-oriented suppression, create conversational dead-ends and fail to address developmental vulnerabilities. CR4T selectively reconstructs unsafe or refusal-style outputs into age-appropriate, guidance-oriented responses, framing adolescent LLM safety as a socio-technical transformation problem rather than a filtering problem. The paper is published under arXiv ID 2605.21609.