Why Accountants Are Right to Be Cautious About AI

What the concerns are actually about

Data security sits at the top of the list for a reason. According to Accounting Today, 83% of accountants are concerned about AI exposing client data — and that concern maps directly to something that’s already happening. In early 2025, one major accounting software provider temporarily suspended its AI assistant after a privacy incident in which a user requesting information about their own invoices was inadvertently shown data belonging to other customers. A flaw in the AI’s data isolation design had surfaced sensitive financial information across unrelated accounts. The issue was fixed quickly, but the episode illustrated something practitioners have been instinctively worried about: when AI systems aren’t built with rigorous data boundaries, the data your firm is trusted to protect can end up somewhere you didn’t intend and can’t fully trace.

Accuracy is the second major concern, and it runs deep in a profession built on getting things right. AccountingWEB’s research describes AI in its current form not as a trusted colleague, but as “a keen but scattershot junior that needs constant supervision.” J.T. Eagan, a clinical accounting professor at Purdue University, tested ChatGPT with one of his own tax questions and described the experience this way: “AI will convince you that the sky is green. It is so convincing. It gave me this response that the mechanics were perfect, but I had to take a step back and say, ‘Well, you’re wrong.’” Generative AI, like ChatGPT, doesn’t catch its own mistakes, and yes, it’ll sound confident no matter what. And while moments like those create lasting distrust around AI, it’s not a reason to avoid it entirely.

And then there’s the concern that gets the least airtime but matters just as much: the relationship question. 63% of accountants worry that AI will erode the personal touch that defines how they work with clients, according to Accounting Today. That fear is understandable since accounting is a relationship business, and clients aren’t just looking for accurate numbers. They want to feel understood by someone who knows their situation, history, and goals. But one accountant quoted in AccountingWEB’s research offered a different perspective: “The future of accountancy is not less human — thanks to AI, it has the potential to be more human.” Her point was that AI handling the repetitive, process-driven work frees practitioners up for exactly the kind of human engagement that defines great client relationships.

The part that gets overlooked

Most accountants are already using AI — and have been for a while. AccountingWEB found that 71% of accounting professionals are using tools like ChatGPT for work: researching tax legislation, drafting emails, summarizing documents.

In most firms, you have AI caution and experimentation happening at the same time.

At industry conferences, AccountingWEB has observed this tension play out in real time, as described in the same survey as above. The sentiment that keeps surfacing is that AI is “terrifying… but terrifyingly brilliant.” One IT professional at a major accounting firm conference spent time walking through his firm’s AI concerns, then immediately pulled out his phone to show the latest tools he was personally experimenting with.

The gap between concern and behavior matters because the consequences are starting to show. Research from Accountancy Age found that 33% of accounting professionals warn that continued reliance on public AI tools could contribute to business failures, and 43% expect a rise in fraudulent or inappropriate claims backed by AI-generated outputs.

Where the profession is drawing the line

When AccountingWEB asked practitioners whether there were services that simply shouldn’t be delivered with AI’s help, the profession was almost evenly split — 55% said AI can be applied across all services with the right oversight, while 45% drew a firm boundary. What’s worth noting is that the 45% weren’t the most skeptical respondents. They were more likely to be active adopters who had simply gotten specific about where AI creates risk rather than value.

When those respondents described where they drew the line, three themes emerged consistently: judgment, empathy, and ethics.

High-level audit work and complex tax planning were flagged as areas requiring human cognition — because AI can process the what and the when, but struggles with the why.

Face-to-face client conversations and sensitive financial discussions were cited as tasks that should stay human-led, with the worry that automating them risks turning a trusted relationship into a transaction.

And ethical reasoning — the kind of moral judgment a regulated profession built on trust depends on — was seen as something that can’t be handed off to a system whose decision-making isn’t fully visible.

This 55/45 split doesn’t describe a profession divided on whether to use AI. It describes two thoughtful approaches taking shape side by side. One group is automating broadly and supervising carefully. The other is automating the heavy lifting while protecting the work that makes an accountant irreplaceable.

Accounting Today’s data adds useful texture here. 55% of accountants would trust AI for research and fact-checking. Only 25% would trust it for direct client communications, and just 9% for advisory services. The degree of trust tracks almost perfectly with the stakes involved.

What to do with the caution

The accounting profession has been more cautious about AI than banking, insurance, or wealth management. Given what the data is showing about AI mistakes, misplaced confidence in public tools, and the gap between concern and behavior at most firms, that caution looks more like foresight than it does like falling behind.

The more useful question isn’t whether to use AI. It’s which tools deserve the trust you’re being asked to extend, and on what basis.

That means getting specific: asking vendors directly where your data goes, whether it’s used for model training, and whether their security environment has been independently audited. It means understanding the difference between public AI models and private ones — because that distinction is where most of the actual data risk lives. And it means matching the tool to the task, which the profession’s own instincts already point toward.

For firms that want to see what deliberate, well-scoped AI adoption looks like in practice — a tool built around a specific workflow, on private AI infrastructure, inside a platform that meets the profession’s security standards — take a look at SmartRequestAI. It’s SmartVault’s AI intake solution that saves firms 60-90 minutes per return.