Absa Group will deploy AI-powered “super agents” to scan its systems for vulnerabilities and shorten patch cycles as it prepares its defences against a new class of frontier AI models capable of finding and exploiting software weaknesses at speed.
Group chief information and technology officer Johnson Idesoh told TechCentral’s Meet the CIO podcast in an interview to be published next week that the bank is actively planning for a world in which tools such as Anthropic’s Mythos give cybercriminals access to capabilities previously available only to nation-state actors.
“The answer I give to our board and to our executive is we will use AI to counter adversaries who use AI,” said Idesoh. “Essentially we will have super agents that use the same technology to do the scanning and vulnerability management ourselves. And then of course, the rate at which a company like ourselves patches these vulnerabilities is also going to have to increase.”
The comments are the most detailed public articulation yet from a major South African bank on how it intends to respond to the emergence of Mythos and similar advanced AI models that are going to test companies’ cybersecurity defences like never before.
Anthropic announced the model on 7 April, describing it as the first AI system whose offensive cybersecurity capabilities were too dangerous for general public release. It previewed Mythos instead to around 40 organisations through a closed initiative called Project Glasswing, which included Amazon, Apple, Microsoft, Google, JPMorgan Chase and the Linux Foundation, among others.
Vulnerabilities
Mythos has been credited with finding thousands of previously unknown zero-day vulnerabilities across major operating systems and web browsers during pre-release testing. The US treasury and Federal Reserve in April convened an urgent meeting with the chief executives of the largest US banks to brief them on the risks posed by the model and similar tools.
Idesoh confirmed Absa has engaged directly with Anthropic and with other Glasswing consortium members. He said the issue has board-level attention at the bank.
Idesoh’s remarks are consistent with the direction other South African banks have taken. Capitec, in its 2026 integrated report, said its AI-driven fraud models had blocked more than 131 000 fraudulent beneficiaries, prevented more than 394 000 scam payments and avoided about R673-million in client losses. The bank has also elevated model risk management to a tier-1 risk and lists AI black-box and agentic AI risks among its emerging risks.
Read: Mythos forces South African banks onto high alert
Digital banking fraud losses in South Africa rose 74% to about R1.9-billion in 2024, according to the SA Banking Risk Information Centre – before frontier AI tools materially shifted the offensive capabilities available to attackers.
The full Meet the CIO interview with Idesoh — which is presented by NTT Data — will be published next week. — © 2026 NewsCentral Media
Get breaking news from TechCentral on WhatsApp. Sign up here.
