Anthropic's Mythos Found Millions of Security Vulnerabilities

Anthropic's Project Glasswing and its newest model, Claude Mythos Preview, are the focus, with the model's ability to identify software vulnerabilities better than most humans. The model is being released to select security researchers and big companies with a $100 million pledge to test and fix open-source software vulnerabilities before public release, due to concerns about potential misuse. The model is a general-purpose model and the next tier above Opus and Sonnet. Anthropic is giving access to Amazon, Apple, Microsoft and others to test code bases and fix vulnerabilities. Internally, Anthropic identified thousands of zero-day vulnerabilities, some dating back 10-20 years. Anthropic's revenue is rapidly growing, surpassing $30 billion, with over 1,000 business customers spending over a million dollars annually.