RFC-020: Governed Autonomy, Symbolic Intent, and D-DNA Evidence — 10. Harm Classes

AIGP Specification › RFC-020: Governed Autonomy, Symbolic Intent, and D-DNA Evidence › 10. Harm Classes

← 9. Autonomous Governance Laws · Section index · 11. Symbolic Governance Language →

10. Harm Classes

AIGP treats human harm broadly.

Governed harm includes but is not limited to:

physical harm
medical harm
privacy harm
security harm
financial harm
legal harm
cognitive harm
psychological harm
dignity harm
agency harm
relational harm
civilian protection harm
protected-status harm

10.1 Cognitive and Psychological Harm

Conversational, companion, therapeutic-adjacent, agentic, and autonomous AI systems may harm humans through interaction trajectories, not merely isolated outputs.

Examples include:

delusion amplification
paranoia reinforcement
self-harm enablement
dependency formation
coercive emotional attachment
false therapeutic authority
reality-boundary erosion
compulsive reassurance loops
social isolation reinforcement
discouragement of human support

AIGP defines a cognitive harm violation as:

An AI system violates governed intent when its interaction trajectory foreseeably degrades a human’s cognitive safety, psychological stability, autonomy, dignity, agency, or reality orientation.

Example governance intent:

governance_intent:
  id: gio.human.cognitive_safety.v1
  purpose: preserve_human_cognitive_and_psychological_safety
  protected_outcomes:
    - reality_orientation
    - autonomy
    - dignity
    - psychological_stability
    - access_to_human_support
    - non_dependency
  prohibited_outcomes:
    - delusion_reinforcement
    - self_harm_enablement
    - paranoia_amplification
    - coercive_attachment
    - false_clinical_authority
    - isolation_from_humans
  priority: human_safety

---

---

← 9. Autonomous Governance Laws · Section index · 11. Symbolic Governance Language →