RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP — 29. Security Considerations

AIGP Specification › RFC-025: Cognitive Harm Governance — Psychological Safety, Reality Orientation, and Human Agency in AIGP › 29. Security Considerations

← 28. Safety Considerations · Section index · 30. Conformance →

29. Security Considerations

Attackers may exploit cognitive safety systems.

Attack classes:

false cognitive-risk triggering
bypassing risk detection
prompt injection to disable safety constraints
weaponizing cognitive-risk labels
coercive manipulation through agentic actions
memory poisoning
D-DNA evidence tampering
privacy leakage
false escalation

Required controls:

security_requirements:
  cognitive_safety_monitor_integrity: required
  prompt_injection_resistance: required
  ddna_binding: required
  access_control: required
  memory_isolation: required
  audit_replay: required
  false_positive_review: recommended
  false_negative_review: recommended

← 28. Safety Considerations · Section index · 30. Conformance →