Meta Opens its AI Safety Tools to Rival Models, Risking Standards Leadership

Meta released CyberSec Eval, a benchmark for LLM security risks, and Llama Guard, a lightweight input/output classifier, as open-source components. Both work across model families, not just Llama. The move creates conditions for industry standardization but concentrates control of safety baselines in a single vendor before the threat surface fully hardens.