Physical AI security · ODA3 Institute
PAI-SF™ v1.0Physical AI Security Framework
Security controls for AI systems whose decisions can produce physical consequences.
Open the authoritative framework ↗Architecture context
From institutional framework to operational practice.
My work on PAI-SF™ focuses on the boundary where AI outputs become physical actions: autonomy limits, command authorization, human override, runtime monitoring, safe-state behaviour, and evidence of enforcement.
Published by ODA3 Institute on 3 August 2026, PAI-SF™ defines 12 domains and 41 controls for physical AI. It expands and supersedes GAISSF™ Domain 9 without deleting the original domain, preserving the relationship between the umbrella framework and the specialized control architecture.
- 01Autonomy boundary
- 02Authorized command
- 03Runtime enforcement
- 04Physical consequence
- 05Evidence and safe-state response