Anthropic's new Claude Mythos AI model has apparently found thousands of vulnerabilities in 'every major operating system and every major web browser, along with a range of other important pieces of software'

Anthropic's new AI model, Claude Mythos, has reportedly identified thousands of high-severity vulnerabilities across every major operating system, web browser, and various other critical software, with 99% of these remaining unpatched. This alarming discovery led Anthropic to launch Project Glasswing, a collaborative initiative involving major tech and finance companies like Apple, Google, Microsoft, and JPMorganChase, aimed at securing the world's most critical software. While many vulnerabilities are merely potential weaknesses, the AI's "Mythos Preview" component has demonstrated the ability to rapidly write exploits for these flaws, a task that typically takes human penetration testers weeks. An example given is an exploit for a 2010 FFmpeg vulnerability stemming from a 2003 bug, which had been overlooked by human reviewers and fuzzers, underscoring the AI's advanced pattern detection capabilities and the urgent need for enhanced security measures. The continuous scanning required, however, incurs significant financial costs due to the demands of large-scale AI server operations.