As AI tools begin matching reconciliations, drafting journal entries and writing variance commentary, finance teams will need to show auditors how those systems reached their work. DataOps president and former auditor Jacqueline DeStefano-Tangorra argues that existing audit controls already provide much of the test: trace the data, reproduce the result, document human review and control who can change or approve an agent’s work.
Watch Desk analysis
What happened
In a Forbes Technology Council article, DeStefano-Tangorra says many public companies could face their first year-end close in which AI makes judgement calls within the process. She notes that the PCAOB does not currently have an audit standard written specifically for AI, while existing standards already cover IT involvement and automated controls in financial reporting.
Her practical advice is to inventory AI use in the close, classify whether it performs or supports a control, preserve inputs and model and prompt versions, and put agent configurations under change management. She also recommends limiting access so an agent cannot both prepare and approve the same transaction, and talking through AI-enabled processes with auditors before fieldwork.
Why it matters
A polished AI-generated summary is only as reliable as the data behind it. If the source extract missed a subsidiary, the commentary can be confidently wrong; if nobody can reconstruct the prompt, model and inputs used at the time, an auditor may struggle to reproduce the result. And a human sign-off is thin evidence if it does not show what was checked or what changed.
These are DeStefano-Tangorra’s professional recommendations, not a new PCAOB rule or a guarantee of what every auditor will demand. But they translate a broad debate about “AI governance” into familiar control questions: where did the data come from, who reviewed the result, and who had permission to alter or post it?
Our read
The useful point is that AI does not make the audit trail optional. It makes the trail more interesting, and potentially much harder to reconstruct after the fact. Finance teams considering AI in the close should start with the controls around data, access, review and change history, rather than treating a model’s output as a self-explaining document. The machine can draft the entry; someone still needs to own the evidence.
What to watch
- Whether companies document AI use in financial reporting before their next audit.
- Whether auditors ask for model, prompt, input and reviewer records when testing AI-supported controls.
- How the PCAOB’s existing technology requirements are applied to AI-enabled finance processes.
Discussion spark: Should AI agents be allowed to draft and post finance entries if a human reviews them afterwards, or must preparation and approval always stay separate?
Sources and evidence
- Your AI Is Now An Internal Control. Your Auditor Will Test It Like One – Forbes (16 October 2026, 11:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.