South African banks are joining their American counterparts by restricting access of third-party AI agents to customer accounts. Absa cautioned customers that they could incur losses if anything unforeseen happens after sharing banking data with an AI agent.
FNB, Discovery Bank, and Capitec have stated that they do not permit third-party AI agents to log into customer accounts. Absa strongly advises customers against providing their credentials to such agents and warns that those who do so may be held liable for any losses.
Meanwhile, FNB is developing a mechanism that will allow approved AI assistants to connect in the future. TechCentral requested information from the banks regarding risks after Axios reported last week on the detection of unauthorized AI agents in the accounts of major American banks; some of these, according to reports, operated independently of user instructions. Reuters followed this report. According to Axios, Morgan Stanley began blocking third-party agents, although a company representative stated that they had not seen 'uncontrolled AI agents accessing our platforms.'
JPMorgan Chase CEO Jamie Dimon stated during the Q3 earnings call on October 13 that the bank 'will not allow any external agent' because 'it is too risky for both clients and JPMorgan.'
As of the publication, Standard Bank and Nedbank did not respond to TechCentral's questions. Dane Liebenberg, Head of AI for Retail and Corporate Banking at FNB, noted: 'FNB currently does not allow third-party AI agents to access customer accounts or conduct transactions for customers.' He added that, in their view, AI assistants will become an important way of serving customers, but access must be secure, regulated, and controlled by the client themselves.
Liebenberg also stated that FNB 'is building a secure foundation for connecting approved AI assistants in the future.' Customers will have to give 'explicit consent for every action,' and identity checks and authentication 'will always take place on secure FNB platforms,' with every interaction being 'tracked and recorded.' The bank will start with informational services and expand functionality 'only after proper safeguards have been implemented.' Furthermore, it is investing in its own AI tools, including the digital assistant NAVi. Liebenberg emphasized: 'Our principle is simple: no autonomy without control.'
Capitec took the most categorical stance, stating that it does not allow third-party AI agents access to customer accounts, which includes 'screen scraping tools.' The bank clarified: 'A customer account can only be accessed by the customer, through Capitec's own secure channels. Currently, an agent cannot access our accounts.' It also stated that Capitec 'would not allow an uncontrolled agent in our clients' accounts.'
However, as Absa points out, an agent who obtains a customer's login data can log in as that customer. Capitec did not specify whether it has encountered customers sharing their login data with agents, but it advises them not to do so. Capitec is building its own tools 'in a secure, regulated, and useful way for our customers,' but has not announced whether it will allow the connection of approved external agents. The bank stated that it is not yet providing AI solutions directly to its customers, although Group CIO Andrew Baker wrote on October 7 about the launch of Scam Checker—'a free AI tool within the Capitec app' to check suspicious links, text, screenshots, and images in the coming weeks.
Joachim Melis, CIO of Absa's Engineering, Services, and Platforms office, stated: 'At Absa, we do not use AI agents to access customer accounts, and we strongly advise customers not to share their banking credentials with any third-party AI agents.' Melis clearly specified that an agent armed with a customer's login data can act on behalf of that customer. He explained: 'These agents do not access customer accounts on their own. Access is only possible when an authorized client provides the necessary credentials or permissions.' He added: 'The main issue is not the technology itself, but the credential exchange. From a security perspective, providing banking credentials to an AI agent is no different from providing them to any other third party.'
Unlike its American peers, Absa does not view agents as a new threat. Melis noted that they 'do not significantly change the bank's current risk profile, as the main security risks are already addressed through established protective measures,' adding that fraud detection systems are designed to identify 'suspicious patterns, including those matching attempts by an unauthorized agent.'
Regarding the development of proprietary agent tools or the creation of a channel for external agents, Absa stated: 'We are studying emerging standards for secure agent interaction and will share further information as the situation develops.'
Discovery Bank reported that it 'currently does not allow third-party AI agents to access customer accounts or initiate transactions.' Instead, the bank is creating its own agent tools through Discovery AI, which allow customers to prepare a payment in the Discovery Bank app using natural language or an image of the bill. The customer must review and approve the payment before it is processed. The bank stated that any loss would be 'investigated based on its individual circumstances, including how access was obtained and whether the payment was authenticated and authorized.'
Who is responsible for losses?
All four banks stated that they would consider loss cases individually. Melis from Absa noted: 'If a customer voluntarily shares such information, and subsequently suffers financial damage, the customer may be liable.' Each case will be assessed under Absa's digital fraud guarantee, which, he said, 'is designed to protect customers who operate responsibly and comply with the bank's security requirements.'
Capitec advised customers not to share their PIN, password, or one-time PIN with 'any third parties, including AI tools or automated agents.' If a loss occurs, 'we assess the circumstances of each case in accordance with our terms and applicable legislation.'
FNB stated that it would 'assess any claimed loss case individually, in accordance with our terms and conditions, as well as applicable regulations.' Liebenberg urged customers never to share their banking login details, PINs, or one-time PINs with anyone, including AI tools or applications.
In the US, such transactions may not be considered unauthorized. Federal rules exclude transfers made by a person to whom the consumer granted access to the account, although the Consumer Protection Bureau argues that account service agreements cannot limit consumers' rights under these rules, and customers deceived into sharing their data remain protected.
Banks are also preparing to work with agents they trust. Visa included South African banks in its Agentic Ready program to prepare for working with AI purchasing agents. On October 15, the Financial Stability Board, which coordinates financial regulation for the G20, finalized its AI guidance for financial institutions, following the June draft that called on banks to rein in AI agents. Among its reasoned practices is 'appropriate and effective human oversight.'
