Tax authorities and AI: risk scoring, audit selection and what triggers a query

Global reference

• Scope: Tax administrations in OECD and partner countries
• Topic: AI in risk assessment, audit selection and fraud detection
• Key sources: OECD Tax Administration 2025; GAO report on IRS AI (March 2026); HMRC annual report 2025-26

The numbers

• OECD: 69% of tax administrations used AI in 2023, up from 9% in 2016, and a further 24% were implementing it, across 58 jurisdictions.
• United States: the IRS had 126 active AI use cases by June 2025, up from 10 in August 2022; 61% were still in development, according to the GAO.
• United Kingdom: HMRC says AI and advanced analytics helped protect and record GBP 10 billion of revenue in 2025-26. Its Connect platform links data from many sources.

How authorities use AI

• Scoring returns for risk and choosing cases for audit.
• Matching returns against third-party data: banks, payroll, property and customs.
• Detecting VAT fraud patterns across networks of companies.
• Reviewing large partnership and mid-market corporate returns, as the IRS reports.
• Monitoring open years continuously, not just at filing.

What triggers a query

• Differences between VAT returns, e-invoices and financial statements.
• Margins or expense ratios far from sector peers.
• Large related-party payments, especially to low-tax countries.
• Customs values that do not match transfer pricing.
• Repeated late filing or amended returns.

What to do now

• Reconcile what you report across taxes: corporate tax, VAT, payroll and customs.
• Run the same analytics on your own data before filing.
• Keep transfer pricing and related-party files current.
• Explain unusual items in the return or a note where the system allows.

Our take

Tax authorities are ahead of many taxpayers on data. Consistency across filings is now the best defence: what the AI flags is the mismatch, not the size of the business.