FRAMEWORK / VERSION 1.0

PAI-SF™

Physical AI Security Framework

A practitioner-first assurance framework for AI-enabled systems that can affect physical state, movement, access, infrastructure, operating environments or human proximity.

12domains
41controls
1.0current release
03 Augpublished 2026

Purpose and scope

PAI-SF™ addresses the assurance chain between digital model behaviour and physical effect: perception → model decision → autonomy boundary → kinetic command → physical effect → monitoring → intervention → recovery.

It applies to robotics, autonomous vehicles, drones and uncrewed systems, industrial automation, medical robotics, smart infrastructure, logistics, embodied agents and other AI-enabled systems that interact with the physical world.

Final Publication v1.0Published 3 August 2026ODA3 Institute

Domains and controls

D1

Sensor & Perception Integrity

Trustworthiness of sensed inputs, perception pipelines and uncertainty handling.

View controls →
D2

Actuator & Kinetic Command Safety

Authorization, validation and constraint of commands that produce physical action.

View controls →
D3

Edge Hardware & Model Attestation

Integrity and provenance of edge compute, firmware, models and trusted execution.

View controls →
D4

Autonomy Boundaries

Explicit limits on delegated authority, mission scope and permissible action.

View controls →
D5

Safe-State & Degraded-Mode Behavior

Predictable transition to bounded or safe operation when confidence or capability falls.

View controls →
D6

Human Override & Intervention

Timely, usable and authoritative human intervention across operating conditions.

View controls →
D7

Runtime Monitoring & Telemetry

Operational visibility into state, decisions, commands, anomalies and interventions.

View controls →
D8

Incident Response & Physical Recovery

Containment, restoration, investigation and learning after physical-AI events.

View controls →
D9

Supply Chain & Update Assurance

Integrity of components, dependencies, maintenance paths and deployed updates.

View controls →
D10

Simulation, Testing & Validation

Evidence across simulation, scenario testing, field validation and operating limits.

View controls →
D11

Functional Safety Interface

Defined handoffs and boundaries between AI security controls and safety mechanisms.

View controls →
D12

Physical Operating Environment

Environmental, spatial, human-proximity and site-specific operating constraints.

View controls →

Publication library

Start with the framework standard and control catalogue. Use the sector, readiness and reference publications for the relevant operational context.

Core framework

NOR-PAISF-005 · Normative standard

Physical AI Security Framework Standard

The primary public framework standard for PAI-SF™ v1.0.

Download standard →
CAT-PAISF-006 · Control catalogue

PAI-SF Control Catalogue

The 41-control public catalogue organized across 12 domains.

Download catalogue →
TCR-PAISF-001 · Technical report

Technical + Compliance Report

Technical foundation, assurance model, limitations and standards context.

Download report →
EXB-PAISF-001 · Executive brief

PAI-SF Executive Brief

Decision-level explanation of the framework, its purpose and adoption boundary.

Download brief →

Start and reference

JSC-PAISF-025 · Schema

JSON Schema Package

Machine-readable public record structures. No scoring thresholds or certification decisions are encoded.

Download schema package →

Sector guidance

Assessment readiness

Using this publication

  • Start with the framework standard for scope, definitions and normative structure.
  • Use the control catalogue to identify domain-level requirements and evidence expectations.
  • Select sector guidance only where its operating context applies.
  • Treat public readiness materials as preparation aids, not assessment outcomes.

Publication boundaries

  • No certification, conformity, safety approval, regulatory recognition or guaranteed effectiveness is implied.
  • No pass/fail threshold, weighting model, maturity trigger, assessor-calibration procedure or certification-decision rule is published here.
  • PAI-SF™ does not replace functional-safety, aviation, medical-device, road, machinery, infrastructure, workplace-safety or other applicable obligations.