Overview
Project Glasswing is a defensive security consortium launched by Anthropic alongside Claude Mythos. It was created to harness the vulnerability discovery capabilities of Claude Mythos for defensive purposes while preventing the technology from being weaponized by malicious actors.
Recent Developments
-
2026-07-17: CNBC sources say company-run Glasswing partner decisions may be superseded by White House gold-eagle greenlighting — disputed by WH official framing (2026-07-18-white-house-gold-eagle-frontier-ai-access)
-
2026-06-24: claude-mythos red-team with US intelligence agencies identified vulnerabilities in classified government systems within hours (2026-06-24-anthropic-mythos-classified-systems-testing)
-
Testing conducted under Glasswing framework with intelligence agency participation
Rationale
Due to the significant potential for misuse if released publicly, Anthropic made the decision not to publicly release Claude Mythos. Instead, the company established Project Glasswing as a controlled access model.
Consortium Members
Key partners include 12 major tech giants and over 40 organizations:
- Cloud Providers: AWS, Google Cloud, Microsoft Azure
- Device Manufacturers: Apple, Samsung
- AI/GPU Companies: NVIDIA
- Security Firms: CrowdStrike, Cisco
- Open Source Foundation: Linux Foundation
Mission
Glasswing enables partners to:
- Scan their own infrastructure for vulnerabilities
- Proactively patch discovered issues in open-source projects
- Contribute to collective defense against emerging threats
- Support independent open-source maintainers
Anthropic’s Commitment
- $100 million in usage credits for the model
- $4 million in donations to open-source security organizations
- Funding to support independent maintainers in patching vulnerabilities
Significance
Project Glasswing represents a new model for responsible AI deployment in security-sensitive applications. It demonstrates that frontier AI capabilities can be leveraged defensively without full public release.