Anthropic unveiled Project Glasswing, a sweeping industry coalition with Amazon, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, Broadcom, Palo Alto Networks, JPMorgan Chase and the Linux Foundation to deploy a new frontier AI model for defensive cybersecurity. The unreleased system, Claude Mythos Preview, has autonomously uncovered thousands of critical zero-day flaws across major operating systems and browsers, including decades-old bugs in OpenBSD and FFmpeg and a privilege-escalation chain in the Linux kernel. Anthropic will provide partners and more than 40 additional organizations access to the model, pledging up to $100 million in usage credits and $4 million for open-source security groups. The company won’t release Mythos broadly, citing misuse risks, and plans to debut enhanced safeguards in a future Claude Opus model. Participants will focus on vulnerability detection, binary testing, endpoint security and penetration testing, with post-preview pricing set at $25 per million input tokens and $125 per million output tokens across major cloud platforms. Anthropic is coordinating with U.S. officials as it pushes for updated security standards—spanning disclosure, software updates, supply-chain protections and automated patching—amid concerns that AI-boosted offensive capabilities could outpace defenders and threaten economic and national security.
Related articles:
Secure by Design, Secure by Default
AI Risk Management Framework (AI RMF 1.0)
Alpha-Omega Project (OpenSSF)
CodeQL: Semantic code analysis for finding vulnerabilities
Apache Software Foundation Security Resources
