Anthropic’s Claude Mythos AI Identifies 271 Vulnerabilities in Firefox

Mozilla has officially integrated patches for 271 security vulnerabilities into the newly released Firefox 150, all of which were identified by Anthropic’s cybersecurity-focused AI model, Claude Mythos. This discovery, made during an early preview of the Mythos model, marks a significant shift in automated threat detection. While the volume of findings is substantial, Mozilla clarified that only three—CVE-2026-6746, CVE-2026-6757, and CVE-2026-6758—required official CVE identifiers. The remaining bugs consist of defense-in-depth gaps, code hardening improvements, and non-exploitable code paths. Firefox CTO Bobby Holley noted that the vulnerabilities were within the range of traditional human discovery, emphasizing that the AI is currently serving as an advanced diagnostic tool rather than an entity uncovering unknowable threats. This collaboration is a key component of Anthropic’s 'Project Glasswing,' an initiative providing limited access to the Mythos model for critical infrastructure partners, including AWS and Microsoft. Data from other participants, such as Palo Alto Networks, indicates that the model can condense a year of traditional penetration testing into less than three…