Project Glasswing
| Source: Anthropic News (community RSS)
Tags: Anthropic, Claude Mythos, Project Glasswing, cybersecurity, vulnerability research, open source security, AI safety
Anthropic's official announcement of Project Glasswing reveals Claude Mythos Preview has already found thousands of high-severity vulnerabilities in every major OS and web browser — and that AI coding capability has crossed a threshold where models can outperform all but the most elite human security researchers.
Details
This is the primary source announcement for Project Glasswing. Anthropic states plainly that Claude Mythos Preview represents a capability threshold: AI models can now surpass all but the most skilled humans at finding and exploiting software vulnerabilities. The model has already discovered thousands of high-severity flaws across every major operating system and web browser during internal evaluation. The consortium launched today includes AWS, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks as founding partners, with more than 40 additional organizations getting access to scan critical and open-source infrastructure. Anthropic is committing up to $100M in usage credits and $4M in direct donations to open-source security organizations. The framing is explicit: Mythos Preview's capabilities will proliferate to other actors — including those not committed to safe deployment — within months to years, not decades. The goal is to put defenders ahead of attackers before that happens. Anthropic characterizes Project Glasswing as 'a starting point,' acknowledging no single organization can solve these problems alone. The model is not available to the public. Access is restricted to named consortium partners and vetted critical infrastructure operators during this phase.