Resources · Answers
What auditors ask about AI in GRC, and what to ask vendors
Auditors ask who approved each AI output, on what evidence, and where that approval is recorded. This article lists five things to ask a vendor before you buy.
AI-native compliance platforms are sold on what the AI does: it detects, drafts, maps and answers questions. This is where the time savings come from. The key question is what happens when an AI proposal becomes a change to the programme. If the software applies the change itself, the audit trail only records what the software chose. It does not record human accountability, and regulators will ask about that.
The approach that holds up in an audit is for the AI to propose and for a person to decide. The AI drafts the mapping, the answer or the classification. Each output shows its source and a confidence score. A person accepts or rejects it, and that decision is what gets recorded. The auditor can then follow each step: the proposal, its citation, the approver and the timestamp.
Ask any GRC AI vendor for five things. Retrieval should use only your organisation's data. Each output should carry a citation. Each output should show its confidence score. Human approval should be recorded as a separate action. The full trail should be exportable. If a vendor cannot show the log of accepted and rejected proposals, the AI's autonomy is a risk to your audit.
Reviewed September 2026.
See this approach in xmodo
A demo shows how xmodo supports the approach in this article, using the frameworks that apply to you.