Read more
But regulators caution against unchecked risk.
Edward Kieswetter says Sars has used machine learning and AI to assess 65% of taxpayers without requiring them to submit a tax return. Picture: Moneyweb
The South African Revenue Service (Sars) has had great success with its tax collection efforts, thanks to artificial intelligence (AI) – and the absence of overly rigid regulation of new technology.
“We are on a path that has no [precedent] and therefore one must balance the development of policy and practice,” said Sars Commissioner Edward Kieswetter on Thursday, when representatives of central banks, the Financial Sector Conduct Authority (FSCA), and the Organisation for Economic Co-operation and Development (OECD) took part in a roundtable discussion about the role of AI in financial systems.
The discussion was a side event of the G20 Finance Track meeting, which is taking place at the Zimbali Resort in KwaZulu-Natal.
“If policy gets too far ahead of practice, we may in fact suppress the experimentation and the exploration of new technology,” Kieswetter notes.
He says Sars has already used machine learning and AI to assess 65% of taxpayers – nine days into the tax filing season – without requiring them to submit a tax return.
“We’ve drawn only on third-party data and used machine learning, algorithms, and AI to perform the assessment and fraud risk detection.”
Twenty years ago, it took Sars six months to assess six million taxpayers.
“This year, they will not have to file. And for the two million who choose to file, they will receive an assessment in under five seconds.”
Within the first nine days of the 2025 tax season, Sars had already paid out R12 billion in refunds and raised R3.5 billion in assessments.
Kieswetter notes that the use of AI also helped Sars prevent R5.5 billion in impermissible refunds.
ALSO READ: Sars makes changes to eFiling for easy use
‘New and complex risks’
Opening the roundtable discussion on Thursday, South African Reserve Bank Governor Lesetja Kganyago warned that while AI holds significant potential for financial innovation and inclusion, it also introduces new and complex risks that make supervision challenging.
“Our mandate tends to converge on key fundamentals to safeguard financial stability, to uphold efficient operation of financial markets, and to deliver consumer protection. These are not abstracts, they are tangible promises we’ve made to our constituents.”
Kganyago added that AI brings “a profound duality – a landscape of immense opportunity interwoven with significant risk”.
On one hand, it can enhance risk detection at both the institutional and system-wide levels, improve consumer credit scoring, and deepen inclusion.
But these same tools can also amplify market shocks and perpetuate bias.
Kganyago emphasised the need for regulators to remain committed to core mandates while adapting to new technologies.
“Technology will change. The tools we use will evolve beyond recognition. But our mandate remains constant – ensuring stability in the value of the monies of the people who are the users of the financial system.”
ALSO READ: Sars records increase in taxpayers who filed returns
AI adoption in South Africa’s financial sector
Katherine Gibson, divisional head at the FSCA, provided new insights from a soon-to-be-published joint survey conducted with the Prudential Authority.
The survey covered nearly 2 000 institutions across banking, payments, insurance, fintech and more, to gauge how AI is being used across South Africa’s financial sector.
“It reveals a sector that is cautiously optimistic about AI’s potential, and, as expected, a range in the pace of adoption,” according to Gibson.
In 2024, global financial sector investment in AI exceeded $50 billion (R892.7 billion), with projections to surpass $150 billion (R2.7 trillion) annually by 2028.
She says locally, fintechs are leading AI uptake, with two-thirds already deploying it. Over 50% of banks have followed suit, while insurance and lending sectors are more reserved, with adoption rates under 10%.
“Banks are investing almost 20 times more than the rest of the sector.”
ALSO READ: Economists say they are confident in Sars
However, she stresses that with innovation comes responsibility.
“We must ensure that hyper-personalisation does not become a byword for hyper-exploitation and that generative AI chatbots are not mis-selling complex products to consumers who do not understand them.”
She warns that some institutions have not implemented transparency safeguards.
“About one fifth of SA institutions reported not using any explainability models or methods for their AI models, which raises questions about transparency and accountability and consumer protection.”
For many institutions, South Africa’s data protection law (Popia) is seen as the top constraint to AI adoption. Gibson disagrees with the notion that regulation is a barrier.
“These are not barriers to innovation, but essential guardrails to ensure tech serves the public good.”
ALSO READ: Sars gets AI to sniff out tax evaders
OECD flags risk of market volatility
OECD Secretary-General Mathias Cormann warns that generative AI brings growing threats of financial fraud, scams and cyber attacks – particularly as global financial markets become increasingly connected through similar AI tools.
“The use of AI in financial markets could also increase market volatility for a significant number of participants that rely on similar AI models, leading to a convergence of trade and feedback loops between models,” Cormann cautions.
The OECD’s most recent risk monitor found that AI use in credit assessments and insurance underwriting can compromise privacy and lead to discrimination. He says competition, rather than concentration, would be key to securing the benefits of AI in finance.
This article was republished from Moneyweb. Read the original here.
