RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP — 26. Replay Semantics

AIGP Specification › RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP › 26. Replay Semantics

← 25. Measurement and Evaluation · Section index · 27. Privacy and Human Rights Considerations →

26. Replay Semantics

D-DNA replay for cognitive harm must reconstruct:

what interaction trajectory occurred
what risk signals appeared
what governance intent applied
what SGL constraints were active
what safety action was taken
whether response behavior changed
whether escalation occurred
whether agentic actions were inhibited
whether privacy controls were applied

Replay MUST support privacy-sensitive review.

Replay SHOULD avoid unnecessary disclosure of full conversation content when hashes, summaries, redacted excerpts, or controlled-access artifacts are sufficient.

← 25. Measurement and Evaluation · Section index · 27. Privacy and Human Rights Considerations →