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

Mozilla discovered 271 security vulnerabilities in Firefox using Anthropic’s Mythos AI model, with almost no false positives.
The key breakthrough was a custom ‘harness’ that guides the LLM through code analysis and uses existing testing tools.
Of the 271 bugs, 180 were rated ‘sec-high’, Mozilla’s highest designation for internally reported vulnerabilities.
Mozilla disclosed details of 12 vulnerabilities, including test cases, to demonstrate the method’s reliability.
Despite skepticism, Mozilla claims AI-assisted vulnerability discovery is a game-changer for software security.