From 2022, Australia's ATO deployed ML to detect a large-scale GST refund fraud scheme: 57,000+ alleged offenders identified, A$2.7 billion in additional suspect refunds prevented, A$123 million recovered, and 126 convictions — with ANAO identifying significant governance gaps.
In May 2022, Australia's Taxation Office (ATO) launched Operation Protego after a rapid surge in fraudulent GST refund claims promoted on social media, in which people set up Australian Business Numbers for non-existent businesses and lodged false Business Activity Statements to claim refunds, with an estimated A$2 billion fraudulently obtained.
Activities
The ATO deployed machine learning models to flag suspicious refund claims in near real time and identify networks of related fraudsters, enabling detection at a scale impossible through manual audit sampling, with AI-driven risk scoring used to prioritise cases for enforcement.
Results
Documented outcomes confirmed by the Australian National Audit Office include prevention of an additional A$2.7 billion in suspect GST refunds, recovery of A$123 million, enforcement action against more than 57,000 alleged offenders, 126 criminal convictions with sentences up to 7 years 6 months, and around A$300 million in penalties and interest levied by July 2025.
Conclusions
The ANAO's 2024-25 performance audit rated fraud detection 'largely effective' in one area but 'partly effective' in three others, finding no fit-for-purpose AI implementation strategy, unclear enterprise-wide AI roles, and no AI-specific risk management arrangements — illustrating both AI's capability to detect large-scale fraud and the governance gaps that can emerge without commensurate oversight.
Implementation
Indicative cost
High (€500k–€5M)
Time to results
Long (> 3 years)
Staffing & skills
ATO investigators using AI-driven risk scoring to prioritise cases, ANAO auditors performing independent governance review
Conditions for success
Real-time or near-real-time fraud flagging at scale
Clear AI governance framework with defined roles and risk management (identified as missing)
Common failure modes
No fit-for-purpose AI implementation strategy
Unclear enterprise-wide AI roles and responsibilities
No AI-specific risk management arrangements (per ANAO audit)
Where it fits
Governance type
national tax administration
Scale
national
Income level
high-income
Data sources
Where this practice's information was retrieved from, and when.
Since 2011, Ireland's Revenue Commissioners have used AI risk scoring (REAP) to target tax-compliance interventions; 290,000 data-analytics-driven interventions in 2023 …
Italy's Agenzia delle Entrate deploys VeRa, an AI algorithm cross-referencing declarations with bank data, property records and electronic payments. In …