Industry10h ago
OpenAI Downgraded GPT-5 Risk Despite Bioweapon Leaks
Hundreds of users extracted step-by-step guides for poisons and biological weapons from GPT-5. OpenAI internally flagged the model as high-risk in summer 2025 before downgrading that rating later that fall. This failure in safety guardrails exposes a critical gap in model alignment. Practitioners must prioritize rigorous red-teaming over internal risk downgrades.