When Anthropic unveiled its new Mythos model in April, it issued a serious warning to software developers. The model was so powerful at identifying software vulnerabilities that it uncovered thousands of high-level bugs, highlighting the urgent need for fixes. Security researchers at Mozilla's Firefox browser explained how this process worked in practice and what Mythos' capabilities mean for software security. This is reported by Techcrunch.com .
Mozilla announced on its blog that Mythos was able to find bugs that had been hidden within the code for over a decade. This is a major leap for AI-based security tools compared to the situation six months ago. While such tools previously distracted security teams with numerous false positives and unnecessary alerts, current agentic systems have the ability to evaluate their work and filter errors.
The results are staggering: in April 2026, Firefox fixed 423 bugs, compared to just 31 in the same period last year. Researchers also identified errors in the analysis of HTML elements that had persisted for 15 years. Mozilla engineer Brian Grinstead noted that the effectiveness of AI in this regard is clearly visible across all industry signals.
The system's ability to identify vulnerabilities in the Firefox sandbox is particularly noteworthy. Typically, finding such complex bugs requires great creativity and a multi-stage process from humans. Even bug-bounty programs offering $20,000 rewards cannot match the volume and quality of the bugs Mythos is finding.
Nevertheless, the Firefox team is not yet fully relying on AI to fix bugs. Although AI can code patches, these codes serve as examples for human engineers rather than being used directly.
Read “Zamin” on Telegram!
