Visa used Mythos to hunt for bugs in its own payment network, then open-sourced the harness that made it possible

Visa tested Anthropic’s Claude Mythos against parts of its global payments infrastructure, where the model linked small weaknesses into exploit chains that might otherwise appear late in penetration testing. The company released the Visa Vulnerability Agentic Harness on GitHub and published a white paper describing its architecture and security practices.

The move shows how major infrastructure operators are beginning to use agentic AI for proactive security, not just defensive monitoring. By open-sourcing the harness, Visa is pushing toward shared standards for AI-assisted vulnerability research in critical systems.