Mozilla says 271 vulnerabilities found by Mythos have "almost no false positives"

Mozilla's use of Anthropic Mythos, an AI model for identifying software vulnerabilities, uncovered 271 Firefox security flaws over two months. The breakthrough was attributed to improvements in the models and a custom "harness" supporting Mythos as it analyzed Firefox source code. Unlike previous AI-assisted detection attempts, Mozilla's work with Mythos resulted in "almost no false positives," thanks to the development of an agent harness that guided the AI model through specific tasks. This harness provided Mythos with access to the tools and processes used by human Mozilla developers, enhancing its accuracy in identifying vulnerabilities.