Claude Mythos Finds 271 Firefox Vulnerabilities - SecurityWeek

Mozilla announced that Anthropic's cybersecurity-focused Claude Mythos AI model discovered 271 vulnerabilities in Firefox 150. The company's security team used targeted prompts with the model alongside internal tools to identify these security flaws. The findings are part of the latest Firefox release, which includes fixes for these vulnerabilities.

Palo Alto Networks provided additional validation, sharing preliminary data from its testing of the Mythos model. The cybersecurity firm reported that Mythos accomplished what it described as 'the equivalent of a year's worth of pentesting in less than three weeks.' This assessment comes from the company's internal evaluation of the model's capabilities for vulnerability discovery.

The announcement follows recent public criticism of Anthropic's marketing approach for Mythos by OpenAI CEO Sam Altman, who characterized it as 'fear-based marketing.' Mozilla's practical application and Palo Alto Networks' testing data provide concrete examples of the model's utility for enterprise security teams seeking to automate aspects of vulnerability discovery.