Claude Mythos Model Identifies 271 Vulnerabilities in Firefox

On April 22, Decrypt reported that Mozilla, the developer of the Firefox browser, announced in a blog post on Tuesday that an early version of Anthropic’s Claude Mythos AI model helped identify 271 vulnerabilities in the Firefox browser during internal testing, which have been fixed this week. The AI model has recently gained attention for its performance in the cybersecurity field. Mozilla stated, “As these capabilities enter the hands of more defenders, many other teams are now experiencing the kind of exhilaration we felt when we first saw the test results. For a hardened target, just one such vulnerability could trigger a red alert by 2025, and the simultaneous emergence of so many vulnerabilities raises the question of whether it is still possible to keep pace.” Mozilla had previously tested another model from Anthropic, which identified 22 security-sensitive vulnerabilities in an earlier version of Firefox. Mozilla also noted that the discovered vulnerabilities were within the capabilities of “elite human researchers.” Mythos, launched in March of this year, is Anthropic’s state-of-the-art model for reasoning, coding, and cybersecurity tasks, representing a new tier beyond its earlier Opus series. Pre-release testing showed it could identify thousands of previously unknown vulnerabilities in major operating systems and web browsers. Access to the system is restricted through a limited project called Project Glasswing, which is only open to specific tech companies like Amazon, Apple, and Microsoft. Reports this Monday indicated that despite the Trump administration’s request for federal agencies to stop using Anthropic technology, the National Security Agency is still running Claude Mythos Preview on classified networks.