Mozilla reports 271 Firefox vulnerabilities discovered via Anthropic Mythos AI model, published May 7

Mozilla used an AI model developed by Anthropic called Mythos to find 271 security flaws in Firefox over two months. Engineers described the results in a post published on Thursday. They credited improvements in the models themselves and a custom harness built to support the AI analysis of the browser source code. The harness wrapped the model and directed it through specific tasks. It supplied instructions such as finding a bug in a file and granted access to tools for reading and writing files and evaluating test cases. The setup ran the model in a loop until the goal was reached. Previous efforts with AI for vulnerability detection frequently generated reports with hallucinated details that demanded substantial human investigation. Mozilla said its approach with Mythos produced almost no false positives. Distinguished Engineer Brian Grinstead noted in an interview that the harness allowed the model to use the same tools and testing pipeline as human developers, including a special Firefox build. A statement from Mozilla's CTO last month said AI-assisted detection meant zero-days are numbered and defenders finally have a chance to win decisively. The company acknowledged skepticism around such claims and provided details on its methods to offer a clearer picture.