Project Glasswing

Project Glasswing is the Anthropic collaboration named in Bytes: Week in Review - Anthropic’s new AI model, a referendum on data centers, and NASA livestreams journey to space. In the episode, it is the mechanism by which Anthropic gives restricted access to [[ClaudeMethosPreview|Claude-Methos Preview]] to more than 40 companies and technology organizations rather than releasing the model to the general public.

The source frames Project Glasswing as a controlled defensive-security rollout. Joanna Stern says the model could help trusted users find long-existing vulnerabilities, but the same capability would be dangerous if it helped offensive actors identify or exploit weaknesses.

Source Position

• The episode names Google, [[JPMorganChase|JPMorgan Chase]], and Cisco as example participants or access recipients.
• The source does not describe the contractual, monitoring, evaluation, or disclosure procedures behind the collaboration.
• The project connects model capability to Frontier Model Access Restrictions: access depends on trusted-user status, not only on payment or public availability.

Name Note

• Project Glassfin is an existing wiki page from a later source that sounds similar and is also linked to Anthropic vulnerability discovery.
• This page preserves the April 10 source’s exact “Project Glasswing” name. Do not merge the two pages unless a later source confirms they are the same project.

Connections

• Anthropic and [[ClaudeMethosPreview|Claude-Methos Preview]] - company and model context.
• AI Cyber-Defense Utility - public-good case for controlled defensive access.
• Frontier Model Release Governance and Frontier Model Access Restrictions - release and access-policy branches.
• Cybersecurity AI Supervision - human oversight pattern for AI cyber agents.