How Anthropic’s Mythos has rewritten Firefox’s approach to cybersecurity

Key Points:

• Anthropic’s Mythos AI model has demonstrated exceptional capability in identifying high-severity software vulnerabilities, uncovering thousands of bugs including some dormant for over a decade in Mozilla Firefox’s code.
• Mozilla’s security researchers report a significant improvement in AI bug-finding tools, with Mythos enabling a dramatic increase in bug fixes—from 31 in April 2025 to 423 in April 2026—by effectively filtering out false positives and assessing its own findings.
• Mythos has uncovered complex sandbox vulnerabilities in Firefox, which are traditionally difficult to detect and highly valued in Mozilla’s bug bounty program, often surpassing the volume of issues found by human researchers.
• Despite advances in AI-assisted coding, Mozilla still relies on human engineers to write and review patches based on AI-generated suggestions, as fully automated bug fixing remains unfeasible.
• The broader cybersecurity impact of AI tools like Mythos remains uncertain; while they enhance defensive capabilities by exposing bugs, there is concern that similar techniques could be exploited by malicious actors, though experts believe the net effect may eventually favor defenders.