The Two Fears Holding Accounting Firms Back From AI — And What to Do About Each

The Two Fears Holding Accounting Firms Back From AI — And What to Do About Each

Here is the paradox buried in Karbon's State of AI in Accounting Report 2026, which surveyed approximately 600 accounting professionals across six continents: 98% of accountants now use AI in some capacity. And 83% of them are still scared of it.

If you're tracking accounting firm AI concerns in 2026, that number — 83% — is the one that tells the real story.

That's not a contradiction. It's a description of exactly where most accounting firms sit right now — somewhere between occasional, cautious AI use and the kind of systematic adoption that would actually change how the firm operates. The distance between those two positions is not capability or cost. It's two specific fears.

Fear #1: AI is going to expose my client's financial data.

Fear #2: AI is going to make my client relationships feel transactional and cold.

Both fears show up in the Karbon data. The data security concern is at 83% — up 7 percentage points from 2025, which tells you that adoption and anxiety are rising together. The human touch concern is at 47% — nearly half the profession worried that the thing clients pay for, the relationship and the judgment, will be eroded by tools that can't provide either.

Both fears are real. Both have specific, answerable responses. This piece works through each one.

---

Fear #1: Data Security — 83% of Accountants Have It

The fear is not paranoia. It's a rational response to a real gap between how AI tools are marketed and how they actually handle data.

Here is what the fear actually looks like in practice: a staff member, trying to be efficient, uploads a client's financial statements into a free-tier AI tool to summarize them. The tool processes the request. The firm has no idea whether that data was used to train a model, stored in a server in another country, or retained after the session ended. If that client later asks what happened to their data during your firm's AI-assisted work, you have no answer.

The fear is legitimate. A 2023 KPMG survey found that 46% of workers admitted to uploading sensitive company information to public AI platforms. That number has almost certainly grown. The problem is not the employees. The problem is that most firms have not told their staff which tools are approved, which categories of data can be shared, and what the acceptable use boundary is.

The reason 83% of accountants still fear this: nobody has given them a checklist that makes the problem manageable. Vendor marketing says "enterprise-grade security." That tells you nothing. The questions that actually tell you something are more specific.

---

Five Questions to Ask Any AI Vendor Before Sharing Client Data

These five questions separate professional-grade tools from risky ones. Ask them before any staff member uses a tool with client data. Ask them again at renewal.

1. Is your platform SOC 2 Type II certified?

SOC 2 Type II is an independent audit that verifies a company's security controls were actually operating effectively over a period of time — not just that the controls exist on paper. It is the baseline credential for professional-grade data handling. A vendor without SOC 2 Type II is not necessarily unsafe, but you need a compelling reason to proceed without it.

2. Do you offer zero data retention — meaning our data is not used to train your model?

Many AI vendors use customer inputs to improve their models unless explicitly opted out. This is disclosed in terms of service, which most users don't read. Zero data retention means your inputs are processed and discarded — not retained, not used for training. Ask for this in writing, not just as a sales claim.

3. Where is data processed — US-only, EU, or global?

This matters for regulatory reasons (GDPR, state data privacy laws) and for practical reasons. If your clients are US-based and their financial data is being processed through servers in jurisdictions with different regulatory standards, you may have an exposure you're not aware of.

4. Will you sign a Data Processing Agreement (DPA)?

A DPA is a contract that specifies exactly how the vendor handles your data, what they can do with it, who has access, and what happens in a breach. Most professional-grade vendors will sign one. Most consumer-grade vendors will not. If a vendor won't sign a DPA, that tells you something about the relationship they're entering into with your data.

5. What happens to stored data if I cancel my subscription?

Data deletion on cancellation should be standard. Ask how long it takes, whether it's automated or requires a support ticket, and whether you can get written confirmation. The answer reveals how seriously the vendor treats data stewardship after the commercial relationship ends.

Professional-grade accounting AI tools — Karbon AI, Intapp Celeste, CCH Axcess Advisor — answer all five. Consumer-grade tools typically can't. The difference is not capability. It's accountability. And accountability is what your firm's professional liability requires.

---

Fear #2: The Human Touch — 47% Worry AI Makes Client Relationships Worse

The 47% concern is worth taking seriously. The accounting profession is built on trust. A firm that automates the relationship away doesn't have a practice anymore. The fear that AI erodes what clients pay for — judgment, context, someone who knows their situation — is not irrational.

But there's a flaw in the framing that makes the fear seem bigger than it is. The fear assumes that the relationship is in the data gathering, the report building, and the analysis. It isn't. The relationship is in the conversation, the interpretation, and the judgment call.

Think about the last time a client called you not because they needed a number, but because they needed someone to help them understand what the number meant, or what to do about it. That conversation — the one where your experience and their context meet — is not something AI can replicate. What AI can do is give you more capacity for it.

The CPA who used to spend two hours building a client's cash flow report now spends thirty minutes. The ninety minutes that freed up goes somewhere. At a firm that has thought about this intentionally, it goes into a longer client call, a more thorough review conversation, or time to reach out proactively about something the client didn't know to ask about. That is more relationship, not less.

Karbon's own data makes this case. The same report that shows 47% concern about human touch also shows that 82% of accounting professionals say AI positively impacts their collaboration, communication, and client relationships. These numbers can coexist because the fear and the outcome are not actually in conflict — the fear is about replacement, and the outcome is about reallocation.

The distinction matters. If you frame AI as "this does the work my clients pay me for," you will erode trust. If you frame it — internally and with clients — as "this frees me to do more of what clients actually value," you get a different outcome.

---

What the Firms That Moved Past Both Fears Actually Did

The firms that have moved from occasional AI use to systematic adoption didn't do it by becoming tech companies. They made three specific operational choices.

They picked specific tools for specific tasks — not "AI for everything."

A 12-person CPA firm doesn't need an AI strategy for every workflow. It needs a decision about which workflows are appropriate for AI-assisted work and which tools are approved for each. Choosing one SOC 2-certified research tool, one SOC 2-certified document drafting tool, and one AI-assisted client communication tool is a manageable starting point. TaxWorld's Ezylia is one example of a no-code tool built for accounting firms with documented data controls — the kind of choice that resolves the vendor question quickly rather than leaving staff to make their own decisions.

They created a one-page AI policy.

Not a 40-page governance document. A one-page policy that answers four questions: Which tools are approved? What categories of data can be shared with each? Who reviews AI-generated outputs before delivery? What is the opt-out process for clients? A policy that exists and is communicated to staff is infinitely better than no policy — it sets a boundary, and boundaries are what make professional use of AI defensible. Firms with a written AI strategy see approximately twice the adoption rate of firms without one, according to Karbon's data. The policy is not just compliance — it's an adoption accelerator.

They told clients.

Not with marketing language. With specific, professional language: "We use AI-assisted tools for [specific task] with human review of all outputs. We use only SOC 2-certified platforms, and your data is never used to train AI models." That statement, delivered once in an engagement letter or onboarding conversation, does more to build trust than silence. Silence reads as either ignorance or evasion. A specific statement reads as competence.

For a deeper look at how to evaluate whether AI investments are generating returns, Measuring AI ROI in Professional Services covers the metrics that matter.

---

The Crossing Report Verdict

Both fears are real. Both have a two-week answer.

The data security concern has a five-question vendor checklist. Run it on every tool your staff uses today. If a tool can't answer all five, restrict it to non-client data until it can. The human touch concern has a reframe: the relationship isn't in the work — it's in the conversation. AI that removes the work doesn't remove the relationship. It creates more room for it.

The firms sitting at 83% concern and not moving are not wrong to worry. They are wrong to stop there. The checklist exists. The policy is a one-pager. The client disclosure is two sentences in an engagement letter. None of this requires a six-month implementation. It requires a Tuesday afternoon.

---

If your firm is still figuring out where AI fits — which tools are safe, which workflows make sense, and how to talk about it with clients — The Crossing Report covers exactly that. Subscribe free — one issue a week, written for professional services firm owners, no vendor spin.