Safety component (AI Act)

How it applies

• Two tests: Either the intended purpose is safety (preventing or mitigating risks to persons or property), or a failure or malfunction would endanger health and safety. One is enough.
• Exclusions: The amended Article 6 excludes systems used solely for user assistance, performance optimization, service efficiency, automation, convenience or non-safety-related quality control — unless their failure or malfunctioning would endanger health or safety. The exception to the exclusion matters: an "optimization" function that can create a hazard when it fails is back in scope.
• Machinery: For machinery, the classification decides whether an AI-enabled machine function needs examination as part of the safety assessment under the Machinery Regulation from January 20, 2027; see the machinery AI timeline.
• Technical documentation: Record the classification for every AI function with its reasoning — intended purpose, failure analysis and why an exclusion applies or not. This link between the risk assessment and the AI inventory is what an assessor will ask for.

Safety component vs. safety function

A safety function is a function of the machine whose failure can immediately increase risk; the AI Act's safety component is the AI system that performs or affects such a function. Calling an AI system a safety component does not by itself say anything about its required reliability — that still comes from machinery risk assessment and functional safety methods.

Compared with the USA, Canada and China

The US and Canada have no AI-specific safety component definition; whether software is safety-related follows from functional safety standards such as IEC 61508 and ISO 13849 and from OSHA and provincial duties. China's machinery standards use the same functional safety concepts, while its AI rules do not classify AI by safety function.