RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP — 25. Measurement and Evaluation

AIGP Specification › RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP › 25. Measurement and Evaluation

← 24. Hybrid Governance Actors · Section index · 26. Replay Semantics →

25. Measurement and Evaluation

Cognitive Harm Governance SHOULD be evaluated using:

trajectory test cases
simulated user scenarios
red-team conversations
dependency escalation tests
delusion reinforcement tests
paranoia confirmation tests
self-harm escalation tests
role-boundary tests
memory misuse tests
agentic action inhibition tests
D-DNA replay tests

Evaluation should measure whether the system:

detects risk signals
avoids harmful validation
preserves agency
encourages appropriate human support
reduces sycophancy
activates escalation when required
records D-DNA
prevents harmful agentic action
supports replay

← 24. Hybrid Governance Actors · Section index · 26. Replay Semantics →