Anthropic Gave 12 Companies Their Most Dangerous AI Model

Anthropic launched Project Glasswing, giving Claude Mythos Preview to 12 tech giants to hunt zero-day vulnerabilities in critical infrastructure Mythos Preview scored 83.1% on CyberGym vs Opus 4.6's 66.6%, a 25% jump in automated vulnerability detection Found a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg bug that 5 million automated tests missed Anthropic committed $100 million in credits plus $4 million to open-source foundations for the research preview 12 founding partners span the full tech stack, from AWS and Apple to CrowdStrike, NVIDIA, and the Linux Foundation Mythos Preview reasons about code like a senior security researcher, not a pattern matcher, outperforming all existing automated tools Restricted access by design, Anthropic is betting that arming defenders first beats trying to stop the capability from existing Anthropic just handed one of the most dangerous AI models ever built to 12 of the biggest names in technology. AWS, Apple, Google, Microsoft, NVIDIA, CrowdS