RFC-021: AIGP-SGL — Symbolic Governance Language for Agentic and Autonomous AI — 19. Cognitive Safety Profile

AIGP Specification › RFC-021: AIGP-SGL — Symbolic Governance Language for Agentic and Autonomous AI › 19. Cognitive Safety Profile

← 18. Force-Capable Systems Profile · Section index · 20. ENFORCE Signal Compatibility →

19. Cognitive Safety Profile

AIGP-SGL also applies to conversational and agentic AI systems that may affect human cognition, psychological stability, reality orientation, dignity, autonomy, or agency.

Cognitive harm may emerge over interaction trajectories, not only isolated responses.

Example:

◎assistant.companion → respond ◇user.statement @conversation.session.993
✕ if response.validates_delusion_as_fact == true
⊘ reinforce:paranoia
⊘ discourage:human_support
! reality_anchor
! encourage:trusted_human_or_professional_support
↯ if self_harm_intent == true
#D-DNA{event:cognitive_safety.constraint.applied, hash:sha256:ab73...}

Human explanation:

The assistant must not validate delusional claims as fact, reinforce paranoia, or discourage human support. It must provide a reality anchor, encourage human or professional support, and escalate if self-harm intent appears.

Required trajectory fields:

cognitive_sgl_required:
  - interaction_window
  - risk_class
  - user_distress_signal_hash
  - response_constraint_hash
  - safety_action
  - escalation_result
  - ddna_event_hash

← 18. Force-Capable Systems Profile · Section index · 20. ENFORCE Signal Compatibility →