Anthropic announced Tuesday the launch of Project Glasswing, a cybersecurity initiative featuring its most advanced AI model, Claude Mythos Preview, which the company says is too dangerous to release publicly. The San Francisco-based AI startup is partnering with twelve major technology and finance companies to identify and patch software vulnerabilities before they can be exploited by malicious actors.

The initiative includes Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks as launch partners. Anthropic has committed up to $100 million in usage credits for the model and $4 million in direct donations to open-source security organizations. More than 40 additional organizations have been granted access to the system.

According to Anthropic, Claude Mythos Preview has already identified thousands of previously unknown vulnerabilities across major operating systems and web browsers. The model discovered a 27-year-old flaw in OpenBSD that allowed remote crashes, a 16-year-old vulnerability in FFmpeg video encoding software, and successfully chained together Linux kernel vulnerabilities to gain complete system control. All discovered vulnerabilities have been reported to maintainers and patched.

The company says it is withholding general release due to the model's powerful cybersecurity capabilities. Newton Cheng, Anthropic's Frontier Red Team Cyber Lead, stated that while such capabilities will likely proliferate to other actors, the company wants to give defenders a head start. On evaluation benchmarks, Mythos Preview scored 83.1% compared to 66.6% for Anthropic's previous best model.

Anthropic has developed a triage system to manage the disclosure of vulnerabilities responsibly, working with professional human reviewers to validate findings before reporting them to maintainers. The company follows a coordinated disclosure framework, typically waiting 45 days after patches are available before publishing technical details publicly.

The announcement comes as Anthropic reported its annualized revenue has surpassed $30 billion, up from approximately $9 billion at the end of 2024, with over 1,000 business customers spending more than $1 million annually. The company also announced a multi-gigawatt compute agreement with Google and Broadcom. Following the research preview period, Claude Mythos Preview will be available to participants at $25 per million input tokens and $125 per million output tokens through various cloud platforms.