Visa lays groundwork for AI payments in South Africa

Visa lays groundwork for AI payments in South Africa


Only 23% of South African consumers would trust an AI agent to complete a purchase on their behalf, according to new research from Visa, even as the payments company prepares technology that would let software shop, compare prices and pay with no human involved.

The finding, released at a media roundtable in Sandton on Tuesday, comes from Visa’s Stay Secure 2026 study, conducted by Wakefield Research between January and February 2026. It covered 5 800 adults aged 18 and older across 17 markets in Central Europe, the Middle East and Africa, including South Africa.

The infrastructure, in other words, is arriving ahead of consumer trust. Lineshree Moodley, country head for Visa South Africa, said local banks had already been enrolled in Visa’s Agentic Ready programme, which prepares financial institutions to process fully autonomous AI-driven transactions. No such transactions have gone live in the country yet.

“Agentic commerce effectively means that you will tell your agent that you want to do something, and it goes ahead, it does the comparisons for you, it does the research for you, and depending on the boundaries you have set, it goes ahead and completes the transaction on your behalf,” Moodley said.

The technology rests on three layers: Visa Intelligent Commerce (a set of APIs that authenticate and personalise agent transactions); the Visa Agentic Ready programme for partner banks and merchants; and the Visa Trusted Agent Protocol, which verifies that a transaction is being initiated by a legitimate agent acting under genuine consumer instruction.

Agentic AI

The most pointed exchanges at the roundtable concerned what happens when an autonomous payment goes wrong. Asked who bears responsibility when an AI agent authorises a fraudulent transaction, Irene Auma, Visa’s head of risk for Eastern Africa, said the same rules governing existing Visa transactions apply. The consumer remains the originating party, whether a human or an agent initiates the payment.

“The payment is still being executed under the instruction of the consumer, through the same financial institution, to a merchant,” Auma said. Where fraud occurs, she added, liability is determined case by case through post-fraud investigation under existing Visa rules.

Read: How agentic AI is reshaping banking in South Africa

What Visa would not say directly was whether agentic commerce makes payments safer. Pressed on the point, Auma said she could confirm the product is secure but stopped short of calling it safer. On how that security is measured, she said Visa assesses transactions that have already been processed.

Visa's Lineshree Moodley
Visa’s Lineshree Moodley

The same AI capabilities that enable agentic commerce are being turned against consumers. Visa told the roundtable that 42.5% of fraud attempts now involve AI and that AI-powered scams grew by more than 1 210% in 2025 – though the company attributed both figures to internal monitoring and did not specify the methodology or geographic scope behind them.

Auma said automated fraud operations had largely replaced human-led ones. “The threat actors are already using AI agents to execute scams. They do not rely on humans to try and call you for your data, send you a fake e-mail. They today have scale already,” she said.

Visa said its AI fraud detection achieves between 92% and 98% accuracy, against roughly 25% for rule-based systems, and that it has invested US$13-billion in fraud-fighting technology overall.

Minors in the crosshairs

Auma warned that children are increasingly being targeted as entry points into the financial system, with fraudsters using gaming platforms and social media to manipulate minors into facilitating transactions. Separately, the study found that 53% of South African consumers who shop on social media had experienced a scam on those platforms.

Read: How AI agents could rewrite the rules of South African banking

TechCentral contacted Mastercard for comment on how it is approaching agentic commerce and AI-driven fraud. The company said a response could take up to two weeks, owing to the internal approval processes required at a global organisation of its scale.  – © 2026 NewsCentral Media