Three accounting executives asked me the same question this month: Our associates are using ChatGPT on engagements. How bad is it?
Processing Content
The market is flooded with hype from big tech and a dozen well-funded startups, each promising autonomous systems that will close the books, review financial statements, and replace workflows that took decades to build. The message is ‘hand it to the AI, and move on.’
We all agree that AI is changing accounting. The PCAOB has been talking about it for a few years. But after seven years of building AI for accounting and audit, one thing rings clear to me: The AI revolution in accounting is real, but removing humans from the loop is the fastest way to destroy the profession.
A failure mode that looks like success
Generative AI applied incorrectly won’t fail with a bang, but with a whimper. It produces polished, authoritative output regardless of whether the underlying data is accurate.
Take a real world example: an AI agent presents audit evidence for a Fortune 500 company. The evidence looks correct, but in reality is slightly off. The opinion is still issued, the financial statements are incorrect, and the 10-K has been filed. The stock trades on numbers nobody verified, compounding chaos ensues. That is not a hypothetical. That is the failure mode.
Unlike a staff auditor who voices uncertainty, generative AI produces confident, well-structured outputs regardless of the accuracy of the underlying content. The result? A hallucinated audit memo built in a block box that looks exactly like a real one but doesn’t have a red flag, an uncertainty signal, or a warning.
When we mapped AI risk across the phases of a PCAOB-governed audit engagement, “evidence evaluation and professional skepticism rated as Critical risk. These are governed by AS 1105 and AS 1000 respectively, standards that exist precisely because the audit profession learned, through hard experience, that trust must be earned through traceable, verifiable human judgment. Autonomous AI, by design, cannot satisfy these requirements. The PCAOB’s Acting Chair
The last layer
AI is absorbing more of the accounting workflow every quarter. The orchestration, the coordination, and the mechanical work are all being claimed by new LLMs.
What AI has not claimed is the moment every audit partner knows. When something feels wrong, they can explain why. A client’s answer that’s technically correct but doesn’t sit right. A number that reconciles but raises a question nobody asked. That instinct is the product of years of practice inside a regulated profession. No prompt replicates it.
As AI takes more of the stack, the human role gets smaller in volume and larger in consequence. Automation still leaves the profession’s hardest work to humans. These are the judgment calls that belong to people, the skepticism that demands experience, and the signature that carries personal liability.
What the profession actually needs
The profession doesn’t need AI that acts on its own. It needs agency, AI that makes its people more powerful.
This distinction speaks to a fundamental truth about audit work. Autonomous AI makes decisions, executes workflows, and produces outputs without a traceable chain of accountability. In a profession where the work is the evidence, that’s a meaningful liability.
The alternative is Auditable AI. AI where every output connects back to a source document, every step is visible, and a human can interrupt, inspect, and override at any point. That is not a lesser version of autonomous AI. It is harder to build. It requires genuine accounting expertise baked into the architecture, not a general-purpose model aimed at audit prompts.
Who will audit the AI?
This is the generational question buried in the efficiency argument. If young accountants only supervise autonomous systems, who develops the judgment required to eventually lead an engagement, sign an opinion, or catch what the machine missed? Professions are living systems. They evolve through practice. Abstract away the practice, and you erode the profession itself.
The firms and regulators that get the human-in-the-loop framework right now will build the template that every other regulated industry, like healthcare, legal, and financial services, will follow. Accounting shouldn’t be viewed as another vertical ripe for AI to disrupt. It’s the canary in the coal mine; the first profession where the consequences of AI hallucination are legally and regulatorily defined.
The audit profession has a once-in-a-generation opportunity to use AI to improve quality, reduce manual work, and give practitioners more time for the judgment that actually defines their value. But that opportunity closes quickly if the profession adopts the wrong version of AI. Speed without accountability is a liability, not progress.
Credit: Source link
