Claude Mythos and the end of software

Anthropic has announced Claude Mythos preview, an unreleased frontier model so capable it has not been made publicly available. The model achieves 78% on SWE-Bench Pro (up from 53% for Opus) and autonomously discovered and exploited zero-day vulnerabilities in major operating systems and browsers, including a 27-year-old OpenBSD vulnerability and a 16-year-old FFmpeg flaw. Because of these dual-use cybersecurity capabilities, Anthropic is restricting access to strategic partners through Project Glass Wing — a coalition including AWS, Apple, Google, Microsoft, Cisco, and others — to patch vulnerabilities before the capability proliferates. Anthropic has committed $100M in usage credits and $4M in donations to open source security efforts. The author argues this represents a turning point where AI models can now chain together complex exploits across all major software, and urges everyone to update their operating systems, browsers, and devices immediately.