RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP — 22. Agentic Action Controls

AIGP Specification › RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP › 22. Agentic Action Controls

← 21. Memory and Personalization Controls · Section index · 23. Autonomous System Interaction Controls →

22. Agentic Action Controls

Cognitive harm is more severe when conversational AI can trigger agentic actions.

Examples:

send messages
make purchases
modify accounts
contact people
delete records
post publicly
call emergency services
change schedules
operate devices

If cognitive risk is active, agentic actions may require additional inhibition.

Example:

◎assistant.agent → send ◇message @session.993
✕ if cognitive_risk >= CR-2 ∧ message_targets_user_family == true
↯ if cognitive_risk >= CR-3
! human_review if action_externalizes_distress == true
#D-DNA{event:cognitive_agentic_action.denied, hash:sha256:...}

---

---

← 21. Memory and Personalization Controls · Section index · 23. Autonomous System Interaction Controls →