Project Glasswing

Overview

Project Glasswing is the Anthropic restricted-access collaboration attached to a cyber-capable model described as Claude-Methos Preview in Marketplace Tech and as Mythos in the new All-In source. Its public meaning in the wiki is controlled defensive access: give trusted organizations time to find, fix, and harden vulnerabilities before similar capability spreads.

Current Profile

The collaboration is framed as a practical test of voluntary frontier-model governance. Rather than releasing a powerful vulnerability-discovery model broadly, Anthropic routes capability to a bounded set of companies and technology organizations, while supporters argue that the arrangement gives defenders a patching window and skeptics question whether the timeline can materially harden the internet.

Key Characteristics

• It is an access-control and coordination mechanism, not merely a model announcement.
• It links Frontier Model Release Governance to concrete remediation work by trusted organizations.
• The bounded sources describe participants or targets as major technology and financial institutions, including Google, JPMorgan Chase, and Cisco in the Marketplace Tech source.
• The All-In source adds a claimed 100-day hardening window and a broader debate over whether voluntary governance can substitute for top-down regulation.
• The sources do not describe the contractual, monitoring, evaluation, or disclosure procedures behind the collaboration.

Evidence

• Access-program evidence: Bytes: Week in Review - Anthropic’s new AI model, a referendum on data centers, and NASA livestreams journey to space says Anthropic gave restricted access to more than 40 companies and technology organizations through Project Glasswing rather than releasing the model to the general public.
• Defensive-window evidence: Anthropic’s $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel’s Influence records a 100-day effort to use advanced AI to find, fix, and harden software vulnerabilities before attackers exploit them.
• Governance evidence: Anthropic’s $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel’s Influence presents Project Glasswing as evidence for market and industry coordination with government, while also recording skepticism that systemic internet patching can happen so quickly.

Qualifications

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 and does not merge the two without confirmation. The All-In source adds discussion but not primary documentation of program scope, monitoring, or remediation outcomes.

What Changed

• Migrated the page to synthesis-v1.
• Added the All-In account of a 100-day hardening effort and the voluntary-governance debate.
• Preserved the Project Glasswing / Project Glassfin naming caution.

Relationships

• Anthropic - organizer of the restricted-access collaboration.
• Claude-Methos Preview - cyber-capable model name attached to the collaboration in the Marketplace Tech source.
• AI Cyber-Defense Utility - public-good rationale for controlled defensive access.
• Frontier Model Release Governance - release-policy frame that Project Glasswing operationalizes.
• Frontier Model Access Restrictions - trusted-user access pattern used by the project.