An algorithm reads the return before a human does. Here is what to tighten.

What actually changed at the IRS

For most of the last decade, IRS audit selection ran on a small set of tools and a shrinking staff. That picture has changed fast. According to CNN Business and a Government Accountability Office report published in March 2026, the agency now runs roughly 126 to 129 active AI use cases, up from about 10 in 2022. AI is no longer a side project at the IRS. It is embedded in the part of the job that matters most to your clients: how a return gets scored and whether it gets pulled.

The practical meaning is simple. A return now passes through algorithmic scoring before a human examiner ever looks at it. The old DIF style high risk scoring has been rebuilt and aimed, and it is pointed at specific kinds of filers. Large corporations, complex partnerships, high wealth individuals, the 400k plus income band, and digital asset and international filers are where the attention sits. If your client list touches any of those, their returns are being read by a model first.

This is not a reason to panic, and the honest framing matters here. The data does not show a flood of new audits. It shows sharper aim.

More targeted, not more numerous

Here is the number that tells the real story. In the first five months of FY2026, IRS enforcement revenue rose about 12 percent, and it did that despite a workforce cut of roughly 25 percent. Fewer people, more money collected. That only happens one way: the cases the agency does open are landing better, because the selection is better.

For a working tax pro, that reframes the whole question. The fear of "more audits" was always a little overblown. The thing that actually changed is the quality of selection. A return that does not reconcile, that contradicts a third party form, or that sits oddly against peers in its category is now more likely to surface, because a model is built to find exactly those signals at scale.

So the defensive move is not to make a client invisible. It is to make their return clean enough that when the algorithm reads it, there is nothing to flag. That is fully inside your control, and it is most of the job already.

What an algorithm rewards and punishes

A scoring model does not get angry or suspicious. It compares. It looks at what the return says against what every connected source says, and it scores the gaps. Once you hold that picture, the priorities for a client file get obvious.

• Reconcile to the third party forms first. W2s, 1099s, K1s, brokerage and digital asset reporting: if the return does not match what the IRS already received, that mismatch is the single loudest signal a model can find. Catch it before they do.
• Substantiate the items that move the number. Large or unusual deductions, basis on digital assets, and anything that swings the liability should have documentation behind it that you could produce on request, not reconstruct in a panic later.
• Keep categories consistent year over year. A return that suddenly looks unlike the same client's prior years, with no explanation, reads as an anomaly. If something genuinely changed, note why in your file.
• Watch the high risk buckets specifically. For partnerships, international positions, and digital asset filers, the scoring is tuned tighter. Those files deserve a second pass.

None of this is new accounting. It is the discipline you already believe in, applied with the knowledge that the first reader is now a machine that only checks consistency.

Where AI helps you on the other side of the desk

There is a fair counterpoint, and it favors you. The same kind of technology that scores returns can also help you prepare cleaner ones. A capable AI assistant is good at exactly the work that prevents flags: cross checking a return against source documents, surfacing line items that do not reconcile, drafting the memo that explains an unusual position, and building a substantiation checklist for a high risk client before filing.

The point is not to hand the model your judgment. It is to use it as a fast, tireless second set of eyes on consistency, so the obvious mismatches never make it onto a filed return. That is the same workflow logic behind the AI tools accountants actually use and the CPA close week AI system: let the machine handle the mechanical cross checking so your attention goes to the calls that need a professional.

One honest caveat. AI gives you scale on the checking, not certainty on the answer. It will miss things, and it will occasionally be confident and wrong. You still read the file, you still verify the items that matter, and you still own what gets filed. The GAO itself flagged that the agency's own AI inventory was incomplete, that there are skills gaps inside the IRS, and that algorithmic scoring carries a risk of unintentional bias. Models are powerful and imperfect on both sides of this.

What to do this quarter

You do not need a project for this. Pull your highest risk files first, the partnerships, the 400k plus clients, the digital asset and international filers, and run each one against the question the algorithm is asking: does everything on this return reconcile to what the IRS already has, and is everything that moves the number substantiated.

Where it does not reconcile, fix it or document why. Where a position is aggressive, write the memo now, while the reasoning is fresh, not eighteen months later under a notice. Then carry that same checklist into the rest of the book. For more on building this around your practice, see how accounting firms run on AI.

The skill under the model

Every filing season there is a new tool, a new IRS capability, a new headline about enforcement. The lesson under all of it is steady. The advantage was never knowing the latest IRS technology. It is having a repeatable method for producing clean, defensible client work, so that whatever the agency points at your return finds nothing.

That method is what protects clients and reputations through every change in how the IRS reads a return. If you want the structured version, Claude AI for Accountants teaches it for working tax and finance professionals, and the two minute course quiz will point you to the right program for your practice.