Anthropic Launches Project Glasswing to Strengthen Global Software Infrastructure Security with Claude Model

Anthropic has launched Project Glasswing, aiming to enhance the security of global critical software infrastructure using the unreleased Claude Mythos Preview model. The program is only accessible to vetted institutions, with participants including Amazon, Apple, Broadcom, Cisco, Google, Microsoft, NVIDIA, and the Linux Foundation.

Anthropic will provide a total of $100 million in model usage credits to help participating institutions detect security issues in their infrastructure. In initial tests, the model identified thousands of high-risk vulnerabilities, including a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg vulnerability that evaded millions of automated tests.

Anthropic commits to releasing a public report within 90 days to disclose vulnerability findings and propose recommendations such as automated patches. Additionally, the company has been in ongoing communication with relevant U.S. government agencies regarding the model's offensive and defensive capabilities.