IRS turns to artificial intelligence to expand audit footprint

The Internal Revenue Service has quietly ushered in a new era of tax enforcement, one powered by machine learning models rather than rows of paper folders and handwritten notebooks. Treasury officials confirmed this month that automation now flags a growing share of the nation's 160 million individual returns, a shift that has already begun reshaping who hears from auditors and why.

For casual filers the change may remain invisible. For anyone with cross-border holdings, freelance income, crypto trades, or a side hustle run through a marketplace app, the maths of getting selected has changed dramatically. Reporters at CAPosts have been tracking filings, court records, and IRS notices to understand where the agency is leaning hardest and what this means for taxpayers who rely on receipts and good faith rather than sophisticated software.

The shift to algorithmic auditing

The IRS has spent several years rebuilding its technology backbone, much of it funded by money Congress provided in 2022 specifically to modernise legacy systems. That rebuild is now mature enough to score returns in real time, comparing declared income against brokerage statements, employer filings, and even bank transaction data pulled under updated information-sharing agreements.

Instead of relying solely on the old Discriminant Information Function, an early statistical model dating back decades, the agency now layers neural networks on top. These newer tools recognise patterns humans cannot, such as specific round-number deductions stacked against income from a particular industry, or gig-economy earnings reported on one platform but quietly absent from a taxpayer's overall numbers.

What's different this filing season

Auditors are no longer limited to reviewing the previous tax year. AI systems help prioritise returns going back several years, looking for inconsistencies that emerge only when you overlay multiple filings side by side. Officers say this year's queue includes a higher proportion of small business owners, content creators, and remote workers who earned income across state lines.

A senior official briefed reporters that the average case load per revenue agent has nearly doubled compared with pre-pandemic norms, even though staffing has recovered only partially. That bump comes directly from the model pre-sorting returns by likelihood of noncompliance, which sends agents fewer dead ends.

The machine learning engine behind the scrutiny

At the core of the new approach is a risk-scoring engine that reads structured forms, parseable attachments, and natural language notes from preparers. It assigns a probability of adjustment and routes high-scoring files into a fast-review workflow that often concludes within weeks rather than months.

Contractors and academics briefed on the platform describe it as part recommender system, part forensic accountant. It weights signals like late 1099 filings by payers, sudden changes in cost-of-goods-sold ratios, and the character of expenses claimed on Schedule C. A return that once sat untouched for years can now surface within hours of submission.

What it means for Australians with US tax exposure

Australia and the United States operate under a tax treaty that prevents most residents from being taxed twice on the same income, but it does not exempt Americans abroad from filing. Dual nationals living in Sydney or Melbourne, along with Australians who own rental property in California or Queenslanders who cashed out shares through a US broker, still receive the same automated scrutiny as anyone stateside.

That matters because Australian superannuation balances, investment properties geared up with negative gearing, and capital gains that qualify for the 50 percent discount under local law can look unfamiliar to algorithms trained on American patterns. A return that flags for a routine review may simply be the system misreading a compliant taxpayer, but defending that position costs time and professional fees.

Melbourne-based expat groups and US-Australia tax forums have lit up with questions about Form 8938 filings and FBAR thresholds, both of which the automated systems cross-check against reported income. Anecdotes from financial planners in Brisbane suggest a steady climb this spring in requests for cross-border reviews, even when the underlying return is clean.

Red flags algorithms are designed to catch

Common triggers include cryptocurrency transactions that aren't reconciled on the return, freelance income reported to the IRS through 1099-K forms but missing from the filer's reported earnings, and home office deductions claimed by employees who have not been reimbursed by their employer for similar costs.

Another category involves passive losses on rental properties, particularly when depreciation is calculated in ways that diverge from software defaults. AI models treat nonstandard calculations as a signal worth examining, even if the taxpayer and their accountant consider the methods entirely legitimate.

Steps to stay ahead of an AI-driven review

Documentation has become the cheapest form of defence. Taxpayers who keep contemporaneous records, especially digital ledgers tied to bank and brokerage accounts, can respond quickly when a notice arrives. So can those who file extensions of practical information, attaching a summary note to unusually complex returns.

Engaging a CPA or enrolled agent familiar with both American and Australian tax obligations is now standard practice for anyone with overseas ties. Perth advisors and Sydney accounting firms have reported rising demand for pre-filing reviews that simulate the kind of automated scrutiny the IRS applies, allowing clients to clear errors before the algorithm flags them.

Catch every fresh CAPosts story on tax, tech, and the markets shaping business across the Pacific by following us on social and subscribing to the daily newsletter at CAPosts.com.