Part 2 – PAP Policymaker Summary
Why PAP Belongs in AI Governance
This is Part 2 of the PAP series.
Readers are encouraged to review Part 1 (PAP Introduction) before continuing.
The Policy Problem
Conversational AI systems can engage users in prolonged, emotionally intense dialogue. While beneficial in many contexts, these interactions may escalate in ways that increase risk—particularly when systems lack clear, enforceable safety interruption mechanisms.
Current governance tools primarily focus on:
- Transparency (system cards)
- Pre-deployment testing (red teaming)
- Risk classification frameworks
- Content moderation rules
- (See AI Safety International glossary for more information on these items)
These approaches describe risks, but they do not define what a system must do when risk emerges during live interaction.
The PAP Solution
The Physiological Aid Protocol establishes a preventive, system-level safety requirement that activates when observable interaction patterns exceed predefined risk thresholds.
PAP requires that systems:
- Temporarily suspend or de-intensify normal conversation
- Clearly notify users that a safety mode is active
- Provide region-appropriate external support resources
- Log activation events anonymously for certified safety review
PAP explicitly does not:
- Diagnose psychological or physiological conditions
- Provide therapy or medical advice
- Monitor or measure human physiology
Why This Matters for Policy
PAP introduces something currently missing from AI governance:
- A clear, auditable safety control
- A non-clinical safeguard that avoids medical claims
- A mechanism that respects user autonomy while reducing systemic risk
By avoiding diagnosis or interpretation, PAP:
- Reduces liability exposure
- Avoids regulatory overlap with healthcare frameworks
- Aligns with established safety-engineering precedent
PAP is therefore suitable for certification standards, regulatory baselines, or legislative adoption.
Part 3 explains how PAP functions conceptually, including escalation logic, thresholds, and its alignment with safety-engineering principles.
➡ Part 3: Full PAP Explanation
© 2025 AI Safety International.
This document may be freely shared, referenced, and adapted for educational, policy, and legislative purposes, provided proper attribution is maintained. No endorsement is implied.