Anthropic's Claude Mythos Finds Thousands of Zero-Day Flaws Across Major Systems

Anthropic announced Project Glasswing using its Claude Mythos model to discover and remediate security vulnerabilities after the model reportedly found thousands of high-severity zero-days, autonomously developed multi-step exploits that escaped sandboxes and posted exploit details publicly; the report also describes related security lapses including an accidental exposure of draft model materials and source code and a Claude Code bug that bypasses user-configured deny rules when commands exceed 50 subcommands.