Five questions to ask before you use AI in audit work
Most ethical slips in audit don't start with bad intentions. They start with a good deal, a busy week, or a tool that seems to know best. AI makes all three more common: vendors compete on price, deadlines shrink because "the tool can do it", and the output reads more confidently than ever.
The Global Internal Audit Standards set out five ethics principles. None of them mentions AI, and none of them needs to. They apply just as much when you use a tool as when you don't. Turned into questions, they make a quick check you can run before you share a file with an AI tool or rely on what it gives you back.
1. What am I sharing, where will it go, and is this tool approved for it?
Confidentiality. Information you hold for an audit may be used only for the professional purpose it was given for. Uploading workpapers to an AI tool sends them to someone else's systems, on someone else's terms. A free trial or a personal account is no exception: the terms may let the vendor keep the data, or use it to improve its model.
Before you share anything, check that the tool is approved for that kind of information, and that the protections you rely on are in the contract, not just in an email from a friendly account manager. If in doubt, try the tool on made-up data first.
2. Is anything swaying my judgement?
Objectivity. AI brings new versions of old threats. You might be asked to audit a model you helped build. A vendor might offer a favour while their product is under review. A dashboard built by the system owner might be the only evidence anyone has looked at.
Objectivity doesn't mean having no history. It means recognising what could affect your judgement, or appear to, and disclosing it before the work begins.
3. Do I understand this tool well enough to rely on it?
Competency. You don't need to know how a model works inside. You do need to know what it was given, what it was asked, and what it might have left out. Tools can quietly cap their results, skip files they can't read, or fill gaps with something plausible. If you can't explain how the output was produced, you can't stand behind it.
4. Have I checked the output against the evidence, all of it?
Due professional care. A polished report is not the same as a complete one. Reconcile record counts and totals to the source. Spot-check items you know independently. Ask the tool directly about limits, such as how many records it processed and which it skipped. Keep the prompt and your checks with the working papers, so a reviewer can see how you know the result is complete.
AI can make you faster. It doesn't make you more competent.
5. Would I stand behind the result if someone pushed back, and is it fair to the people it affects?
Integrity. Pressure to soften a finding hasn't gone away; now a summary tool can soften it for you, without anyone deciding to. Make sure the message the board receives is the one the evidence supports, however it is retold. And when AI is used to judge people, through productivity scores, CV screening or monitoring, ask whether it is lawful and fair, and say so if it isn't.
When you're not sure, don't decide alone
Talk to your Chief Audit Executive, and bring in the people who know: Data Protection, Legal, IT Security. Raising a concern early is almost always easier than explaining a problem later.
See these questions in action
Our course The Auditor's Ethics in an AI Era puts each principle into a story, where you make the call. Five stories, 2.0 recommended CPE hours in Ethics.
About the course