General

Risk assessment covers identification, analysis and evaluation together, done methodically, repeatedly and with stakeholders, using their knowledge and the best information available (ISO 31000:2018, 6.4.1). For AI it is redone whenever the system, its data or its uses change, and it draws on the AI objectives and risk sources of Annexes A and B. The step before 6.4.1 supplies its input, so run both together.

What it asks

Risk assessment covers identification, analysis and evaluation together, done methodically, repeatedly and with stakeholders, using their knowledge and the best information available (ISO 31000:2018, 6.4.1). For AI it is redone whenever the system, its data or its uses change, and it draws on the AI objectives and risk sources of Annexes A and B.

Where generative AI changes the answer

The assessment has to be able to say what the system did with a specific input, which means the corpus, prompt version and model version are part of the record.

Reading it in practice

The general step fixes the order of work: identify, analyse, evaluate. Generative AI tempts teams to skip to controls, because a guardrail is easier to buy than an assessment is to run. Hold the order. A control chosen before the risk is identified cannot be shown to address anything, and an auditor will ask which identified risk it treats.

Who it binds

The AI risk owner, the assessment team and the business owner of each generative AI use. For 6.4.1 the named role is the owner of AI risk assessment records repeated across the life cycle.

Evidence an auditor normally samples

• AI risk assessment records repeated across the life cycle

The test you can run

Compare AI risk assessment records repeated across the life cycle with the use as it runs today; the gap between the two is the finding to record.

Common failures

• One assessment at design with none after retraining

Before the draft publishes

For 6.4.1, run the step once, by hand, on one live generative AI use, and keep the working paper. A method that has never been run on a real use is a template. The first artefact to produce is AI risk assessment records repeated across the life cycle.

Where this sits against the draft

6.4.1 carries the risk management process 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 6.4.1 (general) with this record and note the change in the register.