Anthropic Unveils Project Glasswing: An AI to Hunt Down Critical Cyber Flaws
Anthropic has launched Project Glasswing, an initiative to secure critical infrastructure using its powerful new AI model, Claude Mythos Preview. In just a few weeks, the model has already uncovered thousands of previously unknown vulnerabilities across major operating systems and web browsers.
A Coalition of Industry Leaders
- Key partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks.
- Over 40 additional organizations have been granted access to scan their systems for vulnerabilities.
What is Claude Mythos Preview?
Mythos Preview is a next-generation AI that Anthropic is keeping private for now due to its advanced cyber capabilities. The company plans to first refine its safety mechanisms in the next Opus model before considering a public release for any Mythos-class AI.
Unprecedented Benchmark Performance
- SWE-bench Verified: 93.9% (vs. 80.8% for Opus 4.6)
- Terminal-Bench 2.0: 82% (vs. 65.4% for Opus 4.6)
- CyberGym: 83.1% (vs. 66.6% for Opus 4.6)
Autonomous Vulnerability Discoveries
The model has already demonstrated its power by autonomously identifying critical, long-standing security flaws without any human guidance. All discovered vulnerabilities have since been patched.
- OpenBSD: Uncovered a 27-year-old vulnerability in one of the world's most secure operating systems, allowing an attacker to remotely crash any machine.
- FFmpeg: Found a 16-year-old bug in the popular media library that had gone undetected through five million automated tests.
- Linux Kernel: Independently chained several smaller vulnerabilities into a critical exploit, enabling a standard user to gain full administrative control of a machine.
Commitment to Open-Source Security
Anthropic is backing this initiative with up to $100 million in model usage credits for partners. Additionally, the company is donating $4 million directly to open-source security organizations, including the Linux Foundation’s Alpha-Omega project and the Apache Software Foundation.
Future Access and Pricing
Following the preview period, Mythos Preview will be available to partners via the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. Pricing is set at $25 per million input tokens and $125 per million output tokens.
The 'Glasswing' Philosophy
The project's name is inspired by the Greta oto, or glasswing butterfly. Its transparent wings symbolize both the hidden nature of code vulnerabilities and the transparency Anthropic aims to bring to cybersecurity.
Within 90 days, Anthropic will publish a public report detailing what was found, what was patched, and what industry practices should change to improve security for everyone.