IRS AI audit selection in 2026: What DIF scores actually mean for US expats
The IRS uses computer scoring and newer artificial intelligence tools to help identify tax returns for further review, but its audit process is not a fully automated "AI auditor." Under the IRS's August 10, 2026 AI governance policy, AI that influences whether a taxpayer is audited is presumed to be a high-impact AI use, which brings added testing, monitoring, human oversight, and accountability requirements.
The distinction is relevant because some tax-industry coverage portrays AI as independently deciding who gets audited. IRS and Government Accountability Office records describe a more measured system: computer models help rank or identify returns, and IRS personnel remain involved in most audit-selection decisions.
How does the IRS use AI in audit selection?
The IRS uses AI in parts of its tax-compliance and fraud-detection work to analyze large amounts of data and identify returns or issues with a higher risk of noncompliance. As of June 2025, the IRS inventory contained 43 AI use cases for tax compliance and fraud detection, of which 18 were operational, and 25 were still in development.
According to the GAO's March 24, 2026 review, these systems commonly produce a risk score or recommendation. In most cases reviewed by the GAO, an IRS official then evaluates that output and decides whether to proceed with an audit or investigation.
That is different from saying a computer independently opens an audit.
What is an IRS DIF score?
The Discriminant Inventory Function System, commonly called DIF, is a long-standing IRS computer-scoring system that assigns a numeric score to individual returns and certain corporate returns. A higher DIF score indicates a greater likelihood that an examination could result in a change to the taxpayer's reported income tax liability.
The IRS says its statistical screening compares returns with "norms" for similar returns. Those norms are developed from audits of statistically valid random samples through the National Research Program.
In practical terms, DIF asks whether a return presents a different compliance profile from comparable returns. IRS research describes DIF models as being developed for specific groups of returns rather than applying one universal score across every taxpayer.
A high score does NOT establish that a return is incorrect. It is a selection signal.
What does this mean for US expats?
There is no published IRS rule saying that living abroad automatically produces a higher DIF score. What the IRS does disclose is that DIF compares returns against statistical norms for similar returns, while many expat returns contain combinations of foreign wages, exclusions, foreign tax credits, overseas accounts, and international information forms that differ from a standard domestic wage return. The IRS does not publish the DIF formula, so we cannot say exactly how much weight any one expat-specific item receives.
In other words, an expat return can look less typical without being incorrect. The following 4 features can make an expat return contain more unusual data patterns than a straightforward US W-2 return:
- A large gap between gross foreign income and taxable income. For tax year 2026, a qualifying taxpayer can exclude up to $132,900 of foreign earned income using Form 2555. A US expat who reports $120,000 of foreign wages and qualifies to exclude the full amount could therefore report substantial gross earnings while having little or no US taxable income from those wages. The IRS has not said that this income-to-taxable-income gap automatically raises DIF, but DIF does evaluate returns statistically against comparable return profiles.
- Foreign wages may not have the same US information-reporting trail as domestic wages. A domestic wage return is commonly matched against employer-filed Form W-2 data. Some foreign wages do not come with a US Form W-2, depending on the employer and payroll arrangement, so there can be less US third-party wage information available for matching. The IRS also uses a separate Underreported Income DIF, or UIDIF, score to measure the potential for unreported income. IRS materials do not disclose its formula, and current disclosure rules specifically protect DIF and UIDIF scores from release.
- FATCA can give the IRS a separate foreign-account data trail. Under FATCA, foreign financial institutions or foreign tax authorities can report information about certain US-held accounts, including account balances or values and, depending on the reporting regime, interest, dividends, or other payments. That information can give the IRS another source against which taxpayer reporting may be checked. Public IRS guidance does not, however, say that every difference among FATCA data, Form 8938, FBAR, and Schedule B creates an automatic audit mismatch, and the reporting requirements for those forms are not identical.
- International forms create more reporting points that have to agree with the underlying records. Form 1116, Form 2555, Form 8938, Form 5471, Form 3520, and Schedule C can add income, credits, exclusions, assets, ownership information, or business expenses to a return. The IRS has not publicly stated that filing Form 1116 or another international form automatically increases a DIF score, so that claim would go beyond the available evidence. For self-employed taxpayers, IRS classification guidance does specifically tell reviewers to consider reported income, business expenses, financial status, and whether the return indicates potential unreported income.
None of these items means an expat return is wrong. They mean that an international return can contain more data points, more calculations, and fewer conventional US information documents than a basic domestic return.
TFX's guide to common IRS audit triggers for US expats explains the practical filing issues that can draw IRS attention and the records that can help support a return if questions arise.
What can US expats do to support their return?
The goal isn’t to make an international return look artificially "normal,” but ensure every material figure can be traced to records and that overlapping international filings are consistent where the rules require them to be.
These 4 practices can help:
- Keep evidence for Form 2555 eligibility. If you claim the Foreign Earned Income Exclusion under the physical presence or bona fide residence test, retain records supporting your dates abroad and foreign residence, such as travel records, passport information, leases, and employment documentation.
- Use a consistent and supportable currency-conversion method. The IRS states that it has no official exchange rate and generally accepts a posted rate used consistently. Foreign income is ordinarily translated using the exchange rate in effect when the income is received or accrued, while an annual average rate can be appropriate in some situations, such as income received evenly throughout the year with a relatively stable exchange rate. Form 8938 has its own valuation rules and generally requires Treasury exchange rates for asset values where available.
- Reconcile foreign-account and income reporting before filing. FBAR, Form 8938, Schedule B, and FATCA reporting do not ask for identical information, so differences are not automatically errors. Still, account ownership, balances, interest, dividends, and other reportable amounts should be explainable from the taxpayer's records.
- File every international form required by your facts, even when US tax due is $0. Form 5471, Form 3520, Form 8938, and FBAR have separate filing rules that do not disappear simply because exclusions or credits eliminate US income tax. For example, if more than $5,000 of gross income attributable to specified foreign financial assets is omitted, IRC §6501(e)(1)(A)(ii) can extend the assessment period from 3 years to 6 years. Failure to properly report a required Form 8938 asset can also affect when the statute of limitations closes.
You cannot see your DIF or UIDIF score, and the IRS does not disclose the formulas used to calculate them. What you can control is whether the return, international forms, and supporting records tell the same factual story if the IRS reviews them.
Is DIF itself the IRS's new AI audit system?
No. DIF predates the current generation of IRS artificial intelligence tools by decades. IRS research says DIF was first implemented for return scoring and selection in 1969 and describes it as a supervised machine-learning technique used to predict the likelihood of a material tax change.
Newer IRS AI tools can perform related but distinct functions, such as identifying anomalies, ranking potential compliance cases, or recommending returns or issues for further review.
For that reason, describing the current process as "AI assigns your DIF score and decides whether to audit you" is too broad. DIF scoring, newer AI models, information matching, related examinations, random selection, and human classification can all form parts of IRS examination selection.
Does a human review AI audit recommendations?
Yes, IRS policy requires human oversight for high-impact AI used in audit selection, although not every IRS AI use case operates in exactly the same way. IRM 10.24.1 says AI that influences whether a taxpayer is audited is presumed high-impact, and high-impact projects must provide additional human oversight, intervention, and accountability.
The GAO found that in most tax-compliance and fraud-detection AI systems it reviewed, AI produced a risk score or recommendation, and an IRS official decided whether to pursue an audit or investigation.
There is one clarification. GAO identified two fraud-detection use cases that automated a decision without human review. That finding means it would be too broad to say every IRS AI output always receives the same human review. The better-supported statement is that AI influencing audit selection is subject to formal human-oversight requirements, and most compliance-selection systems reviewed by GAO retained a human decision point.
How much IRS AI is actually in use?
The latest detailed GAO inventory shows that most IRS AI use cases were still in development as of June 2025. GAO reported that 77 of 126 active AI use cases, or 61%, were in development, while the remainder were in operation. This 61% figure comes from the GAO's review of the IRS inventory – not from IRM 10.24.1.
For tax compliance and fraud detection specifically, the June 2025 inventory contained 43 use cases. Of those, 18 were operational, and 25 were in development.
The distinction between the two sources is relevant here.
- IRM 10.24.1 defines the governance structure and lifecycle stages, including pre-deployment, pilot, deployed, and retired status.
- GAO provides the numerical snapshot showing how many projects were in development or operation in June 2025.
The current IRM does not provide a September 2026 count showing how many of those projects have since moved into deployment. Claims that all or most IRS audits are now "AI-driven" therefore go beyond the available public evidence.
Why the "AI auditor" narrative goes too far
The "AI auditor" description goes too far because current IRS policy and GAO findings describe AI primarily as a tool that can generate risk scores, recommendations, or other inputs for compliance work rather than a system that independently conducts the full audit-selection process. For audit-selection uses, IRM 10.24.1 requires high-impact AI controls that include human oversight, intervention, and accountability.
Many sources describe the IRS as an "AI-driven enforcement agency," suggest that "automated audits are here," or present AI as replacing human audit selection. Repetition of that language doesn’t make a fully automated audit process an IRS fact.
The IRS has expanded its use of AI, including in tax compliance and fraud detection. At the same time, GAO's latest detailed inventory found that 61% of IRS AI use cases were still in development as of June 2025, and most compliance and fraud systems it reviewed retained an IRS official at the decision stage.
For taxpayers, the more accurate distinction is between computer-assisted risk selection and the actual examination process.
What should an expat do if selected for an IRS audit?
Handle an audit notice based on the issue and deadline stated in the IRS letter, rather than assumptions about whether an algorithm selected the return. The IRS says it initiates audits by mail, not by an unexpected telephone call.
For an expat responding from abroad, 4 practical steps are useful:
- Verify that the notice is genuine and identify the tax year and issue under examination.
- Calendar the response deadline shown on the letter.
- Gather the records supporting the specific income, exclusion, credit, deduction, foreign account, or entity reporting being questioned.
- Have a tax professional review the response when the examination involves international forms or several tax years.
The selection method does not change the underlying task: the taxpayer must support what was reported on the return.
FAQ
Yes, the IRS uses AI in some audit-selection and compliance work, but available government sources do not show a fully automated system in which AI independently decides most audits. IRM 10.24.1 treats AI that influences audit selection as presumed high-impact and requires additional safeguards, including human oversight.
A higher DIF score indicates that an IRS examination has a higher predicted potential to produce a tax adjustment compared with other returns of the relevant type. It does not prove that the taxpayer made an error.
No, the IRS does not publish enough detail about its DIF formulas to support that conclusion. Foreign income must be reported correctly when required, but there is no published rule saying foreign income or an overseas address automatically raises a taxpayer's DIF score.
No, not according to the latest detailed GAO inventory. As of June 2025, 77 of 126 IRS AI use cases – 61% – were in development, while 49 were in operation. The 61% figure comes from GAO's analysis of the IRS inventory and should not be described as a figure published in IRM 10.24.1.
For most tax-compliance and fraud-detection systems reviewed by GAO, AI generated a score or recommendation, and an IRS official decided whether to proceed. GAO did identify two fraud-detection use cases that made automated decisions without human review, so the answer depends on the specific IRS system.
The governing IRS policy is IRM 10.24.1, IRS Policy for Artificial Intelligence Governance. The current version was transmitted on August 10, 2026, and applies across IRS offices and to contractors and vendors covered by the policy.
Related articles
Discover how AI is transforming IRS efforts to close the growing $688 billion tax gap. Learn about its potential, current uses, and challenges in tax compliance.
Read more
Discover exactly how the IRS finds US citizens abroad who haven't filed: FATCA bank data, FBAR, crypto tracking, AI enforcement & more. 2026 guide.
Read more
Got an urgent IRS email, letter, or call? Learn how to verify real IRS contact, spot scams, report phishing, and protect your tax account.
Read more
