Towards a Physical AI Safety Certification Framework: A Synthesis and Proposal

The deployment of artificial intelligence in physical systems — robots, autonomous vehicles, surgical platforms, and other actuator-driven machines — has outpaced the safety standards meant to govern it. This paper synthesizes seven prior contributions into a proposed certification framework for Physical AI: the Physical AI Safety Certification Framework, or PAS-CF.

The argument proceeds from convergence. Three factors point to the same gap. Regulatory pressure is the first: the EU Machinery Regulation, the EU AI Act, and parallel work in other jurisdictions. The second is technical: common-cause failure analysis in machine-learning-bearing safety channels. The third is empirical: a maturity assessment across the industry. Each factor independently surfaces the same finding — existing functional safety standards do not yet contain AI-specific evaluation criteria.

PAS-CF proposes four such criteria. The first covers AI behavior monitoring under distributional shift. The second covers the presence and integrity of a hardware-layer safety mechanism with separation of fault domains. The third covers common-cause failure analysis with explicit β-coefficient reporting. The fourth covers the audit trail and incident response capability of the system in operation.

The framework augments rather than replaces existing standards. It maps onto IEC 61508, ISO 13849, ISO 13482, ISO 10218, and the emerging ISO/IEC TR 5469. We illustrate application across three example architectures: an industrial cobot with vision-guided picking, an autonomous mobile robot for warehouse logistics, and a surgical robot operating in autonomous mode. The implementation roadmap proceeds in three horizons: voluntary self-assessment (12 months), formal proposal to standards bodies (3 years), and a derived ISO/IEC standard with mandatory third-party certification for high-risk Physical AI (5 plus years).

PAS-CF is offered as a starting point for standards-body deliberation. The paper closes with an open invitation to ISO TC299, IEC TC65, certifying bodies, regulators, industry, and researchers to engage with and improve the framework.