Reference

Glossary

Plain-language definitions of the core terms in Physical AI Safety — the standards, concepts, and vocabulary the field is built on.

Physical AI Safety: The discipline of designing and assuring the safety of physical AI systems — robots, autonomous vehicles, and other machine-learning-driven machines that act in the physical world — at assurance levels comparable to established functional safety. It bridges three traditions: AI Safety, Functional Safety, and Robot Safety. Reference ↗
Functional safety: The part of a system's overall safety that depends on the system operating correctly in response to its inputs — including detecting faults and responding so that risk stays acceptable. It is the domain of standards such as IEC 61508. Reference ↗
IEC 61508: The foundational international standard for the functional safety of electrical, electronic, and programmable electronic safety-related systems. It defines Safety Integrity Levels (SIL 1–4) and a full safety lifecycle, and underpins many sector-specific standards. Reference ↗
ISO 13849 / Performance Level (PL): The machinery-safety standard for the safety-related parts of control systems. It rates each safety function by Performance Level (PL a–e), derived from the system's architecture, component reliability, and diagnostic coverage.
Safety Integrity Level (SIL): A discrete measure of the risk reduction a safety function provides under IEC 61508, from SIL 1 (lowest) to SIL 4 (highest). Each level corresponds to a target probability of dangerous failure. Reference ↗
EU Machinery Regulation 2023/1230: The European Union regulation governing the safety of machinery placed on the EU market, replacing the Machinery Directive 2006/42/EC. It applies in full from 20 January 2027 and, for the first time, explicitly addresses machinery with AI-based, self-evolving behaviour. Reference ↗
Kinetic risk: The risk of physical harm that arises from a machine's movement, mass, and force — the failure space unique to Physical AI, where a software fault can become a bodily injury.
Safe state: A defined system condition in which hazards are controlled — for example, a robot at rest with its actuators de-energised. Functional safety requires a system to reach and hold a safe state when a fault is detected.
Edge safety: Safety logic that runs locally on the device, at the edge, rather than depending on a network or cloud connection — so a machine can still reach a safe state when connectivity is lost.