TLDR Tech / InfoSec
Mozilla says 271 vulnerabilities found by Mythos have "almost no false positives"
Mozilla used Anthropic's Mythos AI model to find 271 Firefox security flaws in two months. The breakthrough came from a custom harness that wraps the LLM, gives it access to Firefox's build tools, and runs it in a loop with clear success signals. When analyzing code for memory safety issues, Mythos crafts test cases against Firefox's sanitizer build. If it crashes, a second LLM verifies the finding. Of the 271 bugs, 180 were sec-high (exploitable through normal browsing), 80 sec-moderate, and 11 sec-low.
Read the full story