Mozilla addressed 423 security flaws in Firefox this April, a massive surge compared to last year's monthly average. This nearly 20-fold increase in bug detection was fueled by Anthropic’s Mythos model, an AI tool that identifies vulnerabilities traditional methods often miss. Among the discoveries was a high-severity flaw that had remained hidden in the code for over 20 years, capable of being triggered by a malicious webpage without any user interaction.
Artificial intelligence is proving particularly effective at identifying sandbox escapes, which are notoriously difficult to find using standard techniques like fuzzing. While manual review remains a cornerstone of their security strategy, Mozilla is leveraging AI to expand coverage and catch deep-seated legacy bugs that have survived decades of manual audits. This shift highlights a new era of proactive security where AI models help clean up technical debt and secure the modern web browsing experience.
