KPMG staff urged to pair AI speed with tax due diligence

The new IRS backdrop

The IRS is no longer a mostly paper-and-human workflow, and that changes the cost of a bad AI answer. Accounting Today’s May 29 AT Think piece framed the issue plainly: AI can speed tax drafting and review, but a smaller, more automated, less forgiving IRS means firms need tighter policies before they let software shape a position. The IRS has already formalized its own AI governance, replacing interim guidance issued on March 11, 2025 with IRM 10.24.1 on February 10, 2026, and extending that policy to employees, contractors, and vendors who develop, procure, use, or monitor AI.

That matters for KPMG because the old definition of due diligence is too narrow. It is no longer enough to check that a return is technically complete; tax teams now need to verify that AI-generated suggestions are accurate, that the underlying data is clean, that a qualified human has reviewed the output, and that the firm can defend the process if the position is challenged. In a firm where audit, controls, and advisory professionals all rely on disciplined review habits, AI should shorten the first draft, not replace professional skepticism.

Why the government side is forcing the issue

The risk is rising because the government side is digitizing at the same time. The IRS’s Small Business/Self-Employed division says it uses digital self-service options so taxpayers and tax professionals can access information, make payments, establish payment plans, respond to notices, and resolve other tax issues, and it reported more than 240,000 taxpayers using automated tools in 2025. That is not just a customer-service upgrade; it is evidence that the agency is building a more standardized enforcement and service environment around automation.

Direct File shows the same shift in sharper detail. After a 2024 pilot in which more than 140,000 taxpayers filed using Direct File, the IRS made it a permanent filing option and expanded it in 2025 to more than twice the number of states for eligible taxpayers. The Taxpayer Advocate Service’s review of the 2024 filing season also said the IRS enabled digital submission of correspondence and responses to notices. For KPMG staff, that means the IRS is moving toward more digital intake, more structured processing, and less tolerance for loose assumptions or weak backup.

What stronger tax due diligence now looks like

In practical terms, a defensible AI-assisted tax file should show three things: the source data, the machine-assisted draft, and the human review that tested and changed it. That trail matters because the real risk is not only a wrong answer, but the inability to explain how the answer was reached if the IRS asks questions later. The IRS’s AI governance rules, along with its newer privacy guidance on AI issued in April 2026, show that the agency is taking documentation, accountability, and data handling seriously on its side as well.

For KPMG teams, that means preserving more than the final memo. Keep the client data set or workpaper extract that fed the analysis, the prompt or query that generated the AI suggestion, the version of the output that was reviewed, the human edits that were made, and the reviewer sign-off that shows who accepted the position. If a question later turns into an IRS exam issue or a partner review issue, that file should let someone reconstruct the logic without relying on memory.

How KPMG staff should use AI without creating exposure

The smartest use of AI in tax is as a speed layer, not a judgment layer. That is especially true during busy season, when compression pressure can make a polished draft look more reliable than it is. For staff and senior associates trying to build credibility on the promotion track, the differentiator is not simply how fast a draft gets out the door; it is whether the work holds up under review, whether the assumptions were challenged, and whether the file can survive scrutiny from a manager, a partner, or the IRS.

That same logic applies across audit, tax, and advisory. AI can help with initial research, summarization, issue spotting, and first-pass drafting, but it should not be allowed to make the final call on a filing position, a disclosure, or a supportable conclusion. The firms that will avoid audit, regulatory, and reputational damage are the ones that treat AI like a junior helper with a very fast keyboard, then keep the senior human accountable for the answer.

The bottom line for KPMG

The message from the IRS is clear: automation is spreading, governance is tightening, and the agency is building a more digital tax environment that leaves less room for sloppy work. For KPMG, the winning model is not AI versus due diligence, but AI plus stronger discipline, with documentation, review, and professional skepticism as the non-negotiable safeguards. In the next phase of tax work, the teams that move fastest will not be the ones that trust the machine most, but the ones that can prove every step.