Claude Mythos Found 271 Firefox Flaws as Mozilla Says AI Matches Top Human Researchers

Claude Mythos Found 271 Firefox Flaws as Mozilla Says AI Matches Top Human Researchers

N
News Editor 01
2026-07-22 16:00:13
Claude Mythos helped Mozilla uncover 271 Firefox vulnerabilities, all patched in Firefox 150. Mozilla said the model has not shown capability beyond elite human researchers, but its scanning speed sharply changes the economics of vulnerability discovery.
FirefoxClaude MythosMozillaCybersecurityAnthropic

Anthropic’s Claude Mythos uncovered 271 Firefox vulnerabilities during a security evaluation with Mozilla, and all of them were patched in Firefox 150. The batch included one bug that had remained in the codebase for 27 years. Mozilla said the first issue appeared within 20 minutes after the model began scanning the Firefox JavaScript engine: a “Use After Free” memory flaw.

Mozilla becomes the first partner to publish Glasswing results

The work was part of Anthropic’s Project Glasswing, announced in April 2026. The idea is to place Mythos Preview in the hands of organizations able to fix security issues before comparable tools reach attackers. Anthropic said the initiative includes 12 major tech companies, among them Apple, Microsoft, Google, and AWS. The company also committed $100 million in support and $4 million in donations to open-source security groups.

Mozilla is one of the first partners to publicly disclose outcomes from the program. According to Firefox’s patch notes, the update closed all 271 vulnerabilities identified during the Mythos Preview assessment. This was not the first collaboration between the two sides. In an earlier review using Claude Opus 4.6, the model submitted 112 reports over two weeks, leading to confirmation of 22 vulnerabilities. Of those, 14 were rated high severity, accounting for nearly one-fifth of all high-severity Firefox fixes in 2025.

Mozilla’s view: same boundary, different speed

Mozilla’s assessment rested on two points that sit side by side. First, the company said it has not found any class or level of vulnerability that human researchers can discover but the model cannot. Second, it also said it has not seen any flaw that elite human researchers would be unable to find. In Mozilla’s reading, AI and top-tier human security researchers are still operating within the same capability boundary.

Mozilla also pushed back on the idea that future AI systems will uncover entirely new categories of vulnerabilities outside current human understanding. Its argument is that software such as Firefox is complex, but not beyond reasoning. The browser is built in modular form, which means humans can still analyze and verify it.

What changes is the math of code coverage

The sharpest shift came from time and scale. Firefox’s codebase contains roughly 6,000 C++ files. Even strong security teams would need long stretches of work to cover large amounts of legacy code in a systematic way. Mythos completed a broad scan in a matter of weeks, raising the pace by dozens of times.

That difference matters because some flaws survive not because humans are incapable of finding them, but because there is never enough time to work through every old component on the list. Mozilla’s tone was cautiously positive. Mythos Preview is still limited to organizations using it for defense, yet Anthropic has also warned that no one knows how long that gap will last.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
400

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.