Anthropic's AI security tool found 271 vulnerabilities in Firefox 150, matching human expert researchers

Anthropic deployed a new AI model called Mythos to audit Mozilla's Firefox browser and discovered 271 security flaws. The company's CTO claims the model performs at the level of top human security researchers—a major step toward AI matching specialized expert work.

Unlike previous security-scanning tools that flag obvious issues, Mythos uses multi-step reasoning (the ability to think through a problem in stages, like a human researcher would) to uncover subtle vulnerabilities that require deep code understanding. This means it catches security risks that simpler automated tools miss.

For Firefox users and Mozilla's security team, this signals a shift in how software gets tested before release—AI can now handle security review work that previously required hiring specialized experts or spending weeks on manual audits. For the broader industry, it raises questions about which types of expert jobs AI can credibly replace and whether companies need fewer human security researchers.