Mozilla used Anthropic’s new Mythos Preview model to scan the unreleased source code of Firefox 150 and flagged 271 security holes before the release went public. That’s a big jump from last month, when Anthropic’s Opus 4.6 model spotted 22 bugs in Firefox 148. Mozilla says having AI run through the code early helped it catch flaws that might otherwise have slipped through until after attackers had a chance to exploit them. Firefox CTO Bobby Holley called the results a turning point for defenders. He didn’t spell out how many of those 271 were critical versus minor, but he made it clear that AI cut down the time and effort needed to spot issues. Holley contrasted Mythos with traditional fuzzing tools and the painstaking work of top security researchers. In their view, a single vulnerability used to demand months of expert focus. Mythos did most of the heavy lifting in a fraction of that time. Anthropic had originally held Mythos back, giving access only to a handful of key industry partners. Skeptics warned that the company might be overhyping routine AI improvements. Mozilla’s experience pushes back on that claim. Whether you see this as a harbinger of AI-driven hacking or a boon for defenders, the Firefox test shows real, measurable gains in automated bug hunting.