Visa deployed Anthropic’s Claude Mythos Preview model as part of Project Glasswing to identify security flaws across its payment network spanning over 200 countries and territories. Within one month, the AI discovered more than 10,000 high or critical vulnerabilities by linking minor weaknesses into complete exploit chains. The company’s zero-trust architecture and network segmentation prevented any of these identified flaws from being exploited during the exercise. Visa open-sourced its vulnerability detection framework VVAH on June 10, 2026, with Anthropic providing $100 million in credits for the project and an additional $4 million specifically for the open-source work.
Source: Read the original article
