Mozilla’s Firefox security team says Anthropic’s Mythos AI model has dramatically accelerated bug discovery in the browser, uncovering a large number of high-severity vulnerabilities — including some that had gone undetected for more than a decade.
In a post published on May 7, 2026, Mozilla’s researchers detailed how Mythos, which Anthropic unveiled in April 2026, has changed their security workflow. Firefox shipped 423 bug fixes in April 2026, compared to just 31 in the same month a year earlier. Mozilla has published details on 12 specific bugs, ranging from a pair of sandbox vulnerabilities to a 15-year-old error in how the browser parses an HTML element.
Particularly notable is Mythos’ ability to find vulnerabilities in Firefox’s sandbox — the browser’s most secure layer. Exploiting a sandbox bug requires a complex, multi-step process in which the model must write a compromised patch and then attack the secured environment using that code. Mozilla’s bug bounty program pays up to $20,000 for sandbox vulnerabilities, the highest reward it offers. Brian Grinstead, a distinguished engineer at Mozilla, said Mythos is finding more sandbox issues than human researchers ever did at comparable volume.
Mozilla’s team credits two factors for the shift: models becoming significantly more capable, and improvements in how agentic systems assess their own outputs to filter out false positives — a persistent problem with earlier AI security tools.
Despite the progress in bug detection, Mozilla is not using AI to write the final patches. AI-generated code serves as a reference for human engineers, but each fix still requires one engineer to write the patch and another to review it. “We have not found it to be automatable,” Grinstead said.
The broader security implications remain uncertain. Anthropic CEO Dario Amodei has suggested the tools could ultimately favor defenders, since there are a finite number of bugs to find. Grinstead offered a more cautious view: “It’s useful for both attackers and defenders, but having the tool available shifts the advantage a little bit to defense. Realistically, nobody knows the answer to this yet.”
Source: TechCrunch