Safety

AI-related objective: The AI system does not cause unacceptable harm to people, property or the environment, with hazards analysed and safe states defined where its failure can cause harm. Used when establishing risk criteria (6.3.4), identifying consequences (6.4.2.6) and evaluating risk (6.4.4). Read safety against the objective the use threatens most.

What it asks

AI-related objective: The AI system does not cause unacceptable harm to people, property or the environment, with hazards analysed and safe states defined where its failure can cause harm. Used when establishing risk criteria (6.3.4), identifying consequences (6.4.2.6) and evaluating risk (6.4.4).

Where generative AI changes the answer

Safety has to be judged against the use: a wrong summary is an annoyance in one setting and a hazard in another.

Reading it in practice

For a generative AI system, safety is judged against the use rather than the model. A wrong summary in an internal draft is an annoyance; the same summary in a clinical or engineering decision is a hazard. The evidence is a use-specific safety analysis that names the decision, the worst credible wrong output and the check that stands between the output and the decision.

Who it binds

The AI risk owner, with the objective owner named in the risk treatment plan. For A.10 the named role is the owner of hazard analysis and safety measures for AI systems that can cause harm.

Evidence an auditor normally samples

• Hazard analysis and safety measures for AI systems that can cause harm

The test you can run

Ask for hazard analysis and safety measures for AI systems that can cause harm and for its last review date; an artefact nobody can date is paperwork rather than a control.

Common failures

• Safety considered only for physical systems

Before the draft publishes

For A.10, pick the two objectives the use most affects, write the test you would apply, and record the result before the draft is published. The first artefact to produce is hazard analysis and safety measures for AI systems that can cause harm.

Where this sits against the draft

A.10 carries the annex a: ai-related objectives part of the draft's published scope, which names risk sources, the stakeholders exposed to them, risk analysis, risk treatment and controls. When the draft publishes, compare its wording on A.10 (safety) with this record and note the change in the register.