Mozilla Patches 271 Firefox Vulnerabilities Found by Anthropic's Claude Mythos

Mozilla said early access to Anthropic's Mythos Preview model helped it identify 271 security vulnerabilities in this week's release of Firefox 150, a substantial increase from the 22 security-sensitive bugs Anthropic's Opus 4.6 model found in Firefox 148 last month. Firefox CTO Bobby Holley wrote in a blog post that the results indicate defenders "finally have a chance to win, decisively" in the battle between cyberattackers and cyberdefenders. Mozilla stated that so far it has found no category or complexity of vulnerability that humans can find that the model cannot. Holley noted that the vulnerabilities could have been discovered through automated fuzzing techniques or by elite security researchers, but using Mythos eliminated the need to concentrate many months of costly human effort to find a single bug in many cases. Mozilla also said the AI was not able to turn up any bugs that a human would not have been able to find given enough time and resources. Anthropic had previously said it was limiting Mythos Preview's initial release to a limited group of critical industry partners. The Mozilla results come amid debate over whether the model represents a significant advance in AI-aided security or is a relatively normal step up in capabilities.