Tag
The paper introduces PsychoSafe, a psychologically-informed refusal framework for large language models that improves refusal quality by 28.1% and resource referral by 46.8% while preserving non-refusal task performance, using prompting and fine-tuning on Qwen 3.5 27B.
OpenAI introduces safety updates to ChatGPT that help it better recognize subtle cues of distress or harmful intent over time in sensitive conversations, enabling more careful responses and de-escalation.
OpenAI announced a €500,000 EMEA Youth & Wellbeing Grant program to fund NGOs and research organizations working on AI safety, literacy, and wellbeing for young people across Europe, Middle East, and Africa. Individual grants range from €25k–€100k, supporting practical tools, harm-prevention programs, and independent research on how AI impacts youth development and safety.