Visa used Anthropic's AI to hunt bugs in payment network and open-sourced the harness

Visa uses Mythos to hunt vulnerabilities and releases tool as open source

Visa directed Anthropic's Mythos model (Claude) at the infrastructure handling billions of daily transactions across more than 200 countries. The goal: uncover hidden weaknesses in the payments stack. The result was a functional exploit chain built from vulnerabilities that previously only surfaced during advanced penetration testing.

Technical and operational impact

The model connected seemingly isolated flaws across different network layers, generating chained exploits that simulate real attacks. Traditional discovery of such chains would take weeks of manual security work. With Mythos, the time dropped to hours. Visa also open-sourced the harness that made the experiment possible, allowing other companies to replicate the approach in their own systems.

Business and growth perspective

For companies operating critical infrastructure, AI-driven security test automation means direct cost savings and faster product release cycles. Instead of waiting for quarterly audits, engineering teams can run continuous scans and fix gaps before they become real exploits. This improves customer retention (trust in the service) and unlocks growth velocity without compromising security.

10Dobro positioning

At 10Dobro, we see the application of AI models in critical processes as a leap in scale and automation. Whether in security or audiovisual production, the principle is the same: using intelligent systems to turn manual bottlenecks into autonomous, ongoing flows. Visa's initiative shows that AI use is not limited to chatbots — it is redefining how companies operate, audit, and protect their digital assets at global scale.