Anthropic Mythos Preview: 83% First-Attempt Exploit Rate, Zero-Days in Every Major OS and Browser

Axios·high signal

Anthropic's Claude Mythos Preview — a withheld frontier model — successfully reproduced vulnerabilities and created proof-of-concept exploits on first attempt in 83.1% of cases, and found thousands of high-severity zero-days across every major operating system and web browser. On expert-level tasks that no model could complete before April 2025, Mythos succeeds 73% of the time. Anthropic restricted release to handpicked Project Glasswing partners (CISA, major tech companies) to find and fix vulnerabilities before broader deployment.