mitre-atlas

Agent Control Plane: The LLM Proposes, It Never Authorizes

Agent Control Plane keeps AI agent authorization outside the model, blocking prompt injection with signed policy and human sign-off for risky actions.