A freelance graphic designer in Austin filed her 2025 return in February, confident the numbers were right. Six weeks later, she got a CP2000 notice from the IRS proposing $4,200 in additional tax. The problem: a 1099-K from a payment app she had forgotten about. She is not alone. The IRS has been expanding its use of artificial intelligence and automated matching tools across its enforcement pipeline, and for the 2026 filing season, those systems are catching discrepancies faster than the agency ever could with staff alone.
That does not mean a robot auditor is reviewing every line of your 1040. But certain patterns, especially mismatches between what you reported and what third parties told the IRS you earned, are more likely than ever to generate a letter, a hold, or a request for documentation. Here is what we know about how the technology works, which red flags are drawing the most attention in 2026, and what you can do to stay ahead of it.
AI at the IRS: What Is Actually Happening
The IRS now uses AI-powered chatbots and voicebots in its authenticated taxpayer environments. These tools handle collection-related interactions: helping taxpayers check balances, set up payment plans, and confirm that documents were received. They operate inside secure, identity-verified sessions tied to specific accounts.
But the AI footprint extends well beyond customer service. The Government Accountability Office has documented how the IRS uses artificial intelligence to identify returns at higher risk of noncompliance, support operational modernization, and prioritize where enforcement resources go. In the same reporting, the GAO flagged institutional weaknesses: skills gaps among staff who interpret AI outputs and concerns about the quality of data feeding those systems.
The backbone of all this is the IRS document matching program, detailed in Internal Revenue Manual section 4.1.27. Programs like the Automated Underreporter (AUR), Business Underreporter (BUR), and Affordable Care Act compliance checks compare the income and deduction figures on your return against third-party information returns: W-2s, every variant of the 1099, and broker statements. When the numbers do not match, the system flags the return. This matching process has existed for decades, but AI is accelerating how quickly the IRS can sort through flagged cases and decide which ones deserve staff time.
The financial incentive is enormous. The IRS estimated the gross tax gap at roughly $688 billion per year for tax years 2017 through 2019, the most recent period with published figures. That gap, the difference between what taxpayers owe and what they actually pay, is large enough to justify significant technology investment even as the GAO has questioned whether the agency can reliably measure the results.
The Red Flags Most Likely to Trigger a Closer Look
You do not need to guess what the IRS is watching for. The document matching programs spell it out. These are the return characteristics most likely to generate automated attention during the 2026 filing season:
Unreported income from any source that files an information return. This is the single biggest trigger. If a client, employer, brokerage, bank, or gig platform sent you a 1099 or W-2 and you did not include that income on your return, the AUR system is built to catch it. The IRS receives copies of every information return filed on your behalf, and the matching is automatic.
1099-K discrepancies. Payment platforms like Venmo, PayPal, and Cash App are required to issue a 1099-K when transactions exceed $2,500 in a calendar year for the 2025 tax year. That threshold, part of a phased rollout the IRS announced after repeatedly delaying the original $600 rule from the American Rescue Plan, means millions of gig workers, online sellers, and freelancers are receiving 1099-Ks for the first time. If the total on that form does not appear somewhere on your return, expect a notice.
Cryptocurrency and digital asset reporting gaps. For the 2025 tax year, centralized crypto exchanges began issuing Form 1099-DA to report digital asset transactions under final IRS regulations published in 2024. This is the first year these forms are in play, and if you sold, exchanged, or disposed of cryptocurrency and the proceeds reported on a 1099-DA do not show up on your Schedule D or Form 8949, the matching system will flag it. The “yes or no” digital asset question on the front of the 1040 adds another data point the IRS can cross-reference.
Mismatches between reported deductions and supporting forms. If you claim mortgage interest but the amount does not match the 1098 your lender filed, or if you report student loan interest that differs from your 1098-E, the system notices. The same applies to tuition credits, health savings account contributions, and retirement plan distributions.
Backup withholding discrepancies. When a payer withholds taxes from your payments because you failed to provide a correct taxpayer identification number, the IRS tracks that withholding. If your return does not account for it, the mismatch gets flagged.
ACA-related reporting gaps. Taxpayers who received advance premium tax credits through the health insurance marketplace must reconcile those credits on Form 8962. Failing to attach the form or reporting figures that conflict with the 1095-A sent by the marketplace is a known trigger for automated holds and notices.
Inflated Schedule C deductions without clear business justification. This is harder for automated systems to catch without a human review, but AI-assisted risk scoring can identify returns where business expense deductions are statistically unusual relative to the reported income and industry. A sole proprietor reporting $50,000 in revenue and $48,000 in expenses will draw more algorithmic skepticism than one with a 30% profit margin.
What the IRS Has Not Disclosed
The IRS has not published the specific algorithms, scoring models, or weighting factors its AI systems use to rank returns by risk. The GAO confirmed that AI plays a role in compliance prioritization, but neither the agency nor the GAO has disclosed how individual return characteristics are scored against each other. Whether a missing 1099-K triggers the same level of scrutiny as unreported brokerage income, for instance, is not spelled out in any public document.
There is also no published false-positive rate for AI-assisted case selection. The GAO raised concerns that staff reviewing AI-flagged returns may lack the training to distinguish genuine noncompliance from system errors, which could mean compliant taxpayers receive unnecessary notices. The IRS has not quantified how often that happens.
And the agency has released no projections tying AI-driven case selection to specific enforcement outcomes for the 2026 filing season. The internal tools, including authenticated chatbot portals, are not accessible to outside researchers in ways that would allow independent evaluation of how well the technology performs. Meanwhile, ongoing staffing changes at the IRS add another layer of uncertainty about how flagged cases are actually handled once the AI surfaces them.
How to Protect Yourself Before and After You File
The single most effective step is also the simplest: before you submit your return, reconcile every information return you received against the figures on your filing. Pull your IRS wage and income transcript if you want to see exactly what the agency already has on file. If a 1099 or W-2 reports income or withholding that does not appear on your return, the document matching programs are designed to catch that gap.
For taxpayers with multiple income streams, organizing records to mirror the forms the IRS receives makes a real difference. If you have three brokerage accounts, make sure your Schedule D and Form 8949 reflect the totals from all three 1099-B forms, not just the ones you remember. Gig workers and freelancers should track income in categories that align with 1099-K and 1099-NEC reporting so the numbers map cleanly.
Already filed and realized you missed something? You can submit an amended return using Form 1040-X. Correcting an error before the IRS contacts you generally results in lower penalties and interest than waiting for a notice. The IRS now accepts electronically filed amended returns, which speeds up processing considerably.
If you do get a notice, know what to expect. The most common outcome of an AUR mismatch is a CP2000 notice: a letter proposing changes to your return based on information the IRS received from third parties. It is not an audit. You have the right to respond, provide documentation, and dispute the proposed changes. Most CP2000 cases are resolved by mail without ever reaching an examiner’s desk.
For taxpayers who interact with IRS chatbots or voicebots during collection or follow-up, treat them as front-end tools, not decision-makers. They can confirm account balances, verify that a payment posted, or provide basic procedural guidance. But substantive disputes about income, deductions, or penalties still go through established IRS procedures, and professional tax advice is worth seeking if the stakes are significant.
Faster Systems, Same Stakes
The rise of AI in tax enforcement has not rewritten the rules of compliance. The IRS still relies on third-party reporting and document matching as its primary detection method, the same framework it has used for years. What has changed is speed. Returns that might have taken months to flag can now be identified in weeks. Cases that might have slipped through due to volume constraints are less likely to be missed.
For taxpayers who report all income shown on their information returns and keep supporting records, the practical impact of AI-assisted enforcement is more likely to be faster resolution of minor issues than an increased risk of audit. The overall individual audit rate remains below 0.5% for most income levels, according to IRS Data Book figures, though filers earning above $1 million face significantly higher scrutiny. But for anyone whose return contains unexplained gaps between what they reported and what the IRS already knows, the probability that an automated system will catch those gaps is higher than it has ever been.
