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

4 hours ago

Mozilla CTO's claim that AI-assisted vulnerability detection will end zero-days and let defenders 'win decisively' was met with skepticism.
Mozilla engineers used Anthropic Mythos AI model and a custom 'harness' to identify 271 Firefox security flaws over two months.
The harness guides the AI by providing instructions, tools like file access and test evaluation, and runs it in a loop until completion.
Unlike previous AI methods that produced many hallucinated bug reports, the new setup resulted in 'almost no false positives.'
The harness integrates with Mozilla's existing developer tools and pipelines, including the Firefox build for testing.

Hasty Briefsbeta