Functional Safety Interface
Defined handoffs and boundaries between AI security controls and safety mechanisms.
Domain controls
The following identifiers, titles, objectives and implementation expectations reproduce the canonical public PAI-SF™ v1.0 Control Catalogue. Evidence requirements, assurance expectations, dependencies, exclusions and conformance relevance remain available in the full catalogue.
Objective: Ensure existing functional safety hazard analysis explicitly includes AI-specific failure modes (drift, adversarial input, sim-to-real gap), not only traditional hardware faults. Requirement: Joint safety/AI-governance review of hazard analysis against an AI-specific hazard checklist.
Objective: Map PAI-SF™ controls to relevant safety integrity level (SIL/ASIL) requirements where applicable, identifying gaps. Requirement: Documented mapping exercise per deployed system class.
Objective: Maintain a living record of where functional safety standards do not yet address AI-driven physical behavior. Requirement: Gap register updated as new gaps are identified through operation or incident.
Objective: Ensure AI-driven hazards not covered by the applicable functional-safety standard are escalated, documented, and approved before deployment or continued operation — not silently absorbed as an unaddressed gap. Requirement: Uncovered AI-driven hazards must be escalated to a named safety authority, documented with compensating controls, and formally approved before deployment or continued operation — this is the enforcement action the gap register (FSI-003) exists to trigger.