Mozilla has identified and resolved 271 vulnerabilities in Firefox using Anthropic’s Claude Mythos Preview model.
Mozilla has been working with Anthropic since February to scan Firefox for security issues. The partnership had already yielded results, but the deployment of Claude Mythos Preview takes the scale to a new level.
From dozens to hundreds of vulnerabilities
Mozilla’s initial collaboration with Anthropic led to 22 fixes in Firefox 148, but Claude Mythos Preview uncovered 271 vulnerabilities in a single evaluation. According to the Firefox team, even one such bug would have been a red flag in 2025. Mozilla states the model is as capable as the best human security researchers, but also notes that the model did not find any vulnerabilities that humans would not have discovered themselves.
Claude Mythos is Anthropic’s latest AI model for detecting vulnerabilities and security issues. While the model can effectively identify hundreds of vulnerabilities, its use also carries risks. Mozilla’s results demonstrate that AI tools can help developers discover many vulnerabilities before attackers exploit them. Conversely, in the wrong hands, this could also cause problems for both software companies and users.