Visa's Mythos Test Reveals AI Security Gap Dangers

Visa's early testing of Anthropic's Mythos AI model, Project Glasswing, has brought critical security implications for advanced AI into sharp focus. The technology team observed how swiftly attackers can identify and exploit vulnerabilities within vital code bases, posing substantial security risks.

This situation underscores the growing enterprise security gaps that malicious actors can leverage. As highlighted by Visa's President of Technology, Rajat Taneja, the increasing autonomy of AI means attacks are becoming self-directed, necessitating autonomous defenses, a capability not yet fully realized, creating a significant global risk.

Threat actors are now equipped with AI agents capable of round-the-clock operation, automating reconnaissance and exploitation phases of cyberattacks at scales beyond human capacity. In response, Visa is building proprietary abstraction layers, observability tools, and data guardrails to fortify its autonomous commerce frameworks.

Furthermore, Visa has launched an open-source, AI-driven security framework. This initiative aims to transform vulnerability discovery and remediation into a structured and repeatable process. Taneja is set to share these insights and technical details at VB Transform 2026 on July 15, in a session titled "Inside Project Glasswing and Mythos: Securing the agentic future today."