Apple and Microsoft join Anthropic in new $100 million AI security project

Anthropic's Project Glasswing is a new cybersecurity initiative, intended to identify and fix vulnerabilities in global software infrastructure. The program includes big players like Amazon Web Services, Apple, Broadcom, Cisco, Google, Microsoft, Nvidia, and the Linux Foundation. These partners will use a new frontier model to scan and harden first-party and open-source systems.

The project is based on Claude Mythos Preview, a specialized model that Anthropic claims has reached a level of coding capability that can surpass most humans at finding software exploits. In preliminary testing, Anthropic says that the model autonomously discovered thousands of high-severity vulnerabilities across major operating systems and web browsers. These included a 27-year-old flaw in OpenBSD and a 16-year-old vulnerability in FFmpeg that had survived millions of previous automated tests.

Anthropic is committing $100 million in model usage credits for project participants. The company is also donating $4 million to open-source organizations, specifically $2.5 million to Alpha-Omega and OpenSSF and $1.5 million to the Apache Software Foundation. However, the company does not plan to make Claude Mythos Preview generally available to the public. Instead, the model will be accessible to vetted organizations via the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Usage for these participants will be priced at $25 per million input tokens and $125 per million output tokens.

Project Glasswing is expected to run for several months - Anthropic has also promised a public report within 90 days. The report will detail the vulnerabilities discovered, alongside listing practical recommendations to improve security practices, such as automated patching and improved supply-chain security. Anthropic stated it has also been in ongoing discussions with U.S. government officials regarding the model’s offensive and defensive capabilities.