Safety Policy¶

Short summary; see also the full Safety Policy reference and the normative specification in SPEC-006.

CRP-Safety-Policy is a request and response header that declares what the protocol should do when the Decision Provenance Engine flags a response.

CRP-Safety-Policy: halt-on CRITICAL; redact-on HIGH PII; warn-on MEDIUM;
                   classify-eu-ai-act MANDATORY; audit-mode CHAIN

Think of it as Content-Security-Policy for AI responses: declarative, transport-layer, non-bypassable by application code.