What AI cannot do in an audit
- Anuj A

- May 25
- 2 min read
AI is already changing how audits run. Document review, sampling, risk identification, and analytical procedures are all getting faster. The direction of travel is clear.
That is not the interesting question. The interesting question is what AI cannot do in an audit, and what that means for how finance leaders should think about control.
AI cannot assess intent. The core risk in any audit is whether someone in the company deliberately misstated something. That question is answered by interviews, by patterns of behavior, by inconsistencies between what people say and what documents show, and by the quiet signals that experienced auditors learn to read. None of that is in the data. AI can flag anomalies. It cannot judge whether an anomaly is a mistake, a process gap, or a deliberate action.
AI cannot exercise professional skepticism. Skepticism is not a set of rules. It is a disposition. A willingness to ask the uncomfortable follow-up question. A refusal to accept a clean explanation when the numbers do not fully support it. A pattern of pressing on the weakest part of a story, not the strongest. AI will eventually simulate this. It will not possess it. That difference matters when the integrity of the numbers is on the line.
AI cannot take responsibility. When the audit opinion gets signed, a named human is personally accountable for that signature. That accountability is what gives the audit its weight in the market. No AI system can be accountable in the sense that matters. The CFO, the audit partner, and the audit committee chair carry that weight by name.
The implication is not that AI does not belong in the audit. It does, and the companies that resist it will fall behind. The implication is that AI belongs in the audit as a tool, not as a party. The professional judgment, the skepticism, and the accountability still sit with humans, and have to be built, evaluated, and invested in as if AI did not exist.
Finance leaders who understand this balance will use AI aggressively in audit and invest more, not less, in the human judgment layer around it.
How does your team's use of AI in audit compare to its investment in the judgment layer that sits over it?


Comments