Fairness

AI-related objective: The AI system treats individuals and groups without inappropriate bias or discrimination, with fairness criteria defined with stakeholders and checked in data, model and outcomes. Used when establishing risk criteria (6.3.4), identifying consequences (6.4.2.6) and evaluating risk (6.4.4). Read fairness against the objective the use threatens most.

What it asks

AI-related objective: The AI system treats individuals and groups without inappropriate bias or discrimination, with fairness criteria defined with stakeholders and checked in data, model and outcomes. 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

Fairness failures in generative AI usually appear as a pattern across many outputs, so a single review of a single answer will not find them.

Reading it in practice

Fairness in generative AI usually appears as a pattern across many outputs rather than a defect in one. The evidence is a test set that reflects the people the output affects, the comparison you ran across groups and what you did about a difference. A single review of a single answer cannot evidence fairness, and an auditor should not accept it.

Who it binds

The AI risk owner, with the objective owner named in the risk treatment plan. For A.6 the named role is the owner of fairness criteria and bias assessment results.

Evidence an auditor normally samples

• Fairness criteria and bias assessment results

The test you can run

Take fairness criteria and bias assessment results and trace it to the decision it supported; if the trail stops before a named person, the control is not evidenced.

Common failures

• Fairness assumed from the absence of protected attributes

Before the draft publishes

For A.6, 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 fairness criteria and bias assessment results.

Where this sits against the draft

A.6 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.6 (fairness) with this record and note the change in the register.