Article by Luis Monzon originally published on MyBroadband, May 9, 2026.
Anthropic’s AI model Mythos, part of its Claude software, represented a clear threat to banks and financial institutions in South Africa.
This is according to Bijan Sanii, CEO of Canadian fraud detection provider INETCO.
“South African banks and financial institutions should be concerned,” Sanii told MyBroadband in an emailed Q&A. “Not because Mythos is necessarily being used in fraud scams and bank fraud schemes today, but because it signals where the cyber threat environment is heading.”
Anthropic, one of the world’s most important generative AI makers, positioned Mythos as an extremely capable AI model designed to identify vulnerabilities in critical software beyond human capabilities. The company decided to release Mythos under heavily restricted access through “Project Glasswing” to prevent the model from falling into the wrong hands.
On 21 April, Bloomberg reported that unauthorised users had gained access to Mythos.
“Even restricted AI tools can leak, be misused or be accessed,” said Sanii. “For South African institutions, the issue is that AI is accelerating the discovery, testing and potential weaponisation of software weaknesses.”
Sanii said he expected other models as powerful as Mythos to emerge in the coming years, which could help attackers map environments, identify weak points, and chain vulnerabilities much faster.
Last week, San Francisco-based security firm Theori employed an “AI hacker” called Xint Code to uncover a high-severity Linux vulnerability that could have allowed attackers to gain a root shell. The flaw affected mainstream Linux distributions using kernels built since 2017 and was discovered after a researcher set the AI to follow an insight into an underexplored bug class in Linux’s crypto subsystem.
“In effect, Mythos is accelerating the shift from a slower cyber world built around periodic reviews and after-the-fact remediation to one where banks need in-flight detection and real-time resilience.”
These risks could be amplified for local financial institutions. While large global technology firms and select banks were granted early defensive access to Mythos, most South African firms were excluded. Sanii said that this could widen the gap between organisations that can test and strengthen systems against AI-assisted attacks and those that must defend without this advantage.
Financial institutions under attack in South Africa
Adumo, one of South Africa’s largest payment processors that processes over R100 billion annually, was the target of a data breach in April that involved an external application connected to its system. That same week, the country’s largest bank by assets, Standard Bank, announced that private client information had been shared online. This followed a major data breach by a threat actor called “Rootboy.” While it was not confirmed whether AI was used in the attacks, cybersecurity firm Check Point Software told MyBroadband that hackers were using AI to aid their attacks in South Africa and worldwide.
“AI-assisted tools can help attackers discover and exploit vulnerabilities much faster,” explained Sanii. “Banks should treat AI cyber risk as a governance and resilience issue, not only an IT issue.”
He said that financial institutions in South Africa should expand red-team and penetration-testing programmes to reflect the growth of AI-assisted attack methods. That meant not only testing individual applications but also how vulnerabilities could be chained across APIs, identity systems, payment environments and, crucially, legacy platforms.
“Then there’s the issue of third parties,” he said. “A weakness outside the bank’s own walls can still become a banking problem if it affects customer access.”
Local banks, insurers, retailers, fintechs and payment providers depend on scores of external vendors, processors, cloud platforms, call centres and software services providers.
“The reported Mythos access issue is a reminder that trusted partners can become an exposure point,” he added.
He said the threat Mythos poses to local financial institutions extends beyond the Anthropic AI model. Instead, its development is a signal that cyberattack capabilities are becoming faster.
“South African banks and companies that adapt now will be much better positioned than those that continue to treat cybersecurity as a back-office technology issue.”
