The Australian Taxation Office (ATO) is using new tech tools to fight tax evasion.

The ATO says artificial intelligence (AI) has helped it bring in more than $530 million in unpaid tax and thwart $2.5 billion in fraudulent claims. 

The agency's innovative deep learning models, natural language processing and other AI tech tools has proven instrumental in identifying tax evasion and underpayments.

The ATO's Deputy Commissioner, Marek Rucinski, says the agency's deep learning models have successfully pinpointed $295 million in superannuation guarantee underpayments. Additionally, natural language models have meticulously combed through leaked documents like the Panama Papers, resulting in the discovery of $242 million owed by tax evaders since 2018.

Rucinski emphasised the tangible impact of AI during The Australian Financial Review Government Services Summit, claiming; “We’ve got clear evidence that AI makes an impact, that AI generates results and financial gains... AI is real, it’s here, and it’s fast-evolving as well”

Utilising AI's analytical prowess, the ATO has uncovered invaluable insights within its massive datasets that would have otherwise eluded human detection. 

Since its adoption of AI models in the 2018 fiscal year, the ATO has achieved an alleged 90 per cent success rate in identifying potential superannuation underpayments by employers. 

Moreover, the ATO has incorporated gradient-boosting machine learning models to identify evolving patterns of Goods and Services Tax (GST) fraud. These models build on previous iterations to improve accuracy.

The ATO claims that all AI applications have stringent decision controls and human oversight, emphasising responsible deployment. 

The agency has stressed that AI does not replace human judgement but rather empowers staff by automating routine tasks, enabling them to focus on matters demanding human insight.