Why it matters

Dangerous capability evals shape safety commitments. Understanding shapes safety.

Advertisement

The architecture

Categories: CBRN + cyber + autonomy + persuasion.

Structured probes.

Gate deployment via RSP / PF.

Dangerous capability evalsCategoriesCBRN + cyber + autonomyStructured probesexpert-designedGatingRSP / PF thresholdsFrontier labs publish RSP/PF thresholds + evaluations; open evals emerging
Dangerous capability evals.
Advertisement

How it works end to end

Anthropic RSP (Responsible Scaling Policy).

OpenAI Preparedness Framework.

DeepMind Frontier Safety Framework.