How should professional scepticism be applied when using AI?
Professional scepticism should be applied to AI output in the same way it is applied to other audit evidence: with a questioning mind, attention to possible bias and a willingness to investigate information that contradicts the expected conclusion.
AI creates some specific traps. Fluent language can make weak support appear convincing. A system may favour the information it can retrieve easily, overlook missing evidence or reproduce assumptions embedded in the methodology or prompt. Auditors should not lower their challenge because the result is well structured.
Practical scepticism includes checking the cited evidence, asking what information is absent, considering alternative explanations, reviewing exceptions and understanding where the system has inferred rather than observed. Teams should also avoid using AI only to confirm an existing view.
Well-designed software can support scepticism by surfacing contradictory evidence, showing confidence and making the evidence trail easier to inspect. It cannot exercise professional scepticism on behalf of the auditor. That remains a human responsibility and a core part of audit quality.
References: IAASB professional scepticism resources · IAASB overreliance on technology guidance