Anthropic developed Mythos, an AI model that discovers and generates zero-day exploits at 72.4% success rate, vastly outperforming Claude Opus 4.6. Rather than public release, Anthropic created Project Glasswing, providing preview access exclusively to 12 industry partners including AWS, Apple, Google, and Microsoft for defensive security research.
ModelsFEATURED
Anthropic: All your zero-days are belong to Mythos
Anthropic's Mythos model generates zero-days at 72.4% success but stays private—shared only through Project Glasswing's 12-partner program to contain exploit weaponization.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline
Tags
models
/// RELATED
SafetyApr 7
Anthropic Teams Up With Its Rivals to Keep AI From Hacking Everything
Anthropic partners with 40+ rivals including Microsoft, Apple, and Google via Project Glasswing to privately red-team Mythos Preview and patch vulnerabilities before release, establishing industry-wide coordinated disclosure for AI security.
ModelsApr 7
Anthropic's Project Glasswing - restricting Claude Mythos to security researchers - sounds necessary to me
Anthropic gates Claude Mythos—which discovers thousands of high-severity OS and browser vulnerabilities—to security researchers via Project Glasswing before broader release.