Partnering with Mozilla to improve Firefox’s security

In collaboration with Mozilla, the AI model Claude Opus 4.6 demonstrated its advanced capability to identify and report high-severity security vulnerabilities in complex software like Firefox, discovering 22 vulnerabilities in February 2026 alone. This AI-assisted approach, involving the use of task verifiers to ensure the reliability of findings, significantly accelerates the process of detecting and fixing vulnerabilities, as evidenced by the rapid identification and submission of 112 unique reports to Mozilla, several of which were incorporated into Firefox updates. Despite its proficiency in finding vulnerabilities, Claude Opus 4.6 was less effective at exploiting them, highlighting a current advantage for cybersecurity defenders. This collaboration underscores the importance of using AI as a tool for improving software security, while also raising concerns about potential misuse if AI models become adept at creating exploits. Anthropic plans to expand its cybersecurity efforts and encourages developers to strengthen their software defenses, stressing the need for vigilant monitoring and updating of AI capabilities to prevent potential threats.