Question 1

What is Physical AI Safety?

Accepted Answer

Physical AI Safety is 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, none of which alone addresses the failure modes that arise when machine-learning components drive physical actuators.

Question 2

How is Physical AI Safety different from AI Safety?

Accepted Answer

AI Safety largely concerns the behaviour of software models — alignment, bias, hallucination, and misuse — where a failure is informational. Physical AI Safety concerns systems whose outputs move mass and exert force, where a model failure can become a physical injury. It therefore inherits the assurance methods of functional safety and robotics in addition to those of AI Safety.

Question 3

How is Physical AI Safety different from robot safety?

Accepted Answer

Traditional robot safety and functional safety were built for deterministic, rule-based automation whose behaviour is predictable and can be exhaustively verified. Physical AI Safety extends them to machines whose behaviour is learned and probabilistic, where conventional methods of verification and certification do not fully apply.

Question 4

What safety standards apply to physical AI and robots?

Accepted Answer

The core standards include IEC 61508 (functional safety of electrical, electronic, and programmable electronic systems), ISO 13849 (safety-related parts of control systems, rated by Performance Level), and the robot standards ISO 10218 and ISO/TS 15066 for collaborative robots. In the European Union, the Machinery Regulation (EU) 2023/1230 applies in full from 20 January 2027 and, for the first time, explicitly addresses machinery with AI-based behaviour.

Question 5

Is there a certification for Physical AI Safety?

Accepted Answer

There is no single certification called 'Physical AI Safety.' In practice, manufacturers certify against the established functional-safety and machinery standards — such as IEC 61508, ISO 13849, and ISO 10218 — and, in the EU, must conform to the Machinery Regulation. How those frameworks should be applied to learning-based, probabilistic behaviour is still an open question the field is actively working on.

Question 6

What is a Safety Integrity Level (SIL) or Performance Level (PL)?

Accepted Answer

A Safety Integrity Level (SIL 1–4) specifies how dependable a safety function must be under IEC 61508, with SIL 4 the most stringent. The machinery standard ISO 13849 uses a parallel scale called Performance Level (PL a–e). Both express, in standardised terms, how much a safety function reduces risk.

Question 7

What are examples of physical AI systems?

Accepted Answer

Examples include autonomous mobile robots in warehouses, collaborative robot arms working alongside people, self-driving vehicles, delivery and inspection drones, agricultural and construction robots, and humanoid robots. What they share is machine-learning-driven control of real-world motion and force.