- Jibin Rajan Varghese/
- News/
- Released AILuminate v1.0, an industry-standard AI risk and reliability benchmark/
Released AILuminate v1.0, an industry-standard AI risk and reliability benchmark
Released AILuminate v1.0, the first comprehensive industry-standard benchmark for AI-product risk and reliability, from the MLCommons AI Risk and Reliability working group.
The benchmark scores how a system responds to prompts meant to elicit dangerous or undesirable behavior across 12 hazard categories, from violent crime and weapons to privacy, hate, and specialized advice. We shipped an assessment standard, hidden and practice prompt sets, a response evaluator, and a five-tier grade from Poor to Excellent, so model providers, integrators, and policymakers can talk about safety in the same language.