// PROJECT GLASSWING
Project Glasswing
Anthropic's largest cybersecurity initiative — arming defenders with Claude Mythos Preview to find vulnerabilities before attackers do.
12
Founding Partners
$100M
Compute Credits
1000+
Zero-Days Found
$4M
OSS Donations
Project Glasswing
Securing critical software for the AI era
// OVERVIEW
Why Glasswing Exists
Claude Mythos Preview's cybersecurity capabilities far exceed any existing AI model. It can fully autonomously discover and exploit zero-day vulnerabilities — no human involvement required. This means the barrier for attackers has been fundamentally lowered.
Anthropic's response isn't to hide the model — it's to put it in the hands of defenders. Project Glasswing gives vetted partners early access to Mythos Preview so they can patch critical software before attackers gain equivalent capabilities.
This is the first model Anthropic has published a System Card for without making it generally available. Their assessment: Mythos's cyber capabilities are too powerful for unrestricted public release at this time.
// FOUNDING PARTNERS
11 Industry Leaders
Spanning cloud, endpoint security, silicon, finance, and open-source infrastructure — the critical nodes of the global digital ecosystem.
Amazon Web Services 
Apple 
Broadcom 
Cisco 
Google 
JPMorgan Chase 
Linux Foundation 
Microsoft 
Nvidia 
Palo Alto Networks // VULNERABILITY DISCOVERY
Thousands of Zero-Days, Fully Autonomous
Mythos Preview has identified thousands of previously unknown critical vulnerabilities across every major operating system and browser — including bugs that have existed for decades.
17yr
Age
FreeBSD NFS
CVE-2026-4747
27yr
Age
OpenBSD
CVE-2026-*
16yr
Age
FFmpeg
CVE-2026-*
20yr
Age
Linux Kernel
CVE-2026-*
CVE-2026-4747: A 17-year-old remote code execution vulnerability in FreeBSD's NFS implementation. Mythos discovered, analyzed, and wrote a working exploit with zero human intervention — the world's first fully autonomous AI zero-day exploit chain.
$100M Compute Credits
Allocated to founding partners for large-scale vulnerability scanning and patch validation. At Mythos Preview pricing ($25/$125 per million tokens), this represents massive security audit capacity.
$4M Open-Source Donations
Direct funding to the open-source security ecosystem.
// TIMELINE
Key Milestones
2026.04.08
Project Glasswing Launches
Mythos Preview released simultaneously. 12 founding partners receive full access.
Within 90 Days
First Public Research Report
Partners will publish initial vulnerability findings and remediation guidance.
6-18 Months
Broader Access Assessment
Depending on efficiency improvements and safety progress, access scope may expand.
Available Platforms
Claude API · Amazon Bedrock · Google Cloud Vertex AI · Microsoft Azure AI Foundry
Glasswing FAQ
chevron_right Can individual developers join Glasswing?
Not currently. Glasswing is limited to vetted organizations, primarily founding partners and early access participants. Anthropic says access will expand as safety measures mature.
chevron_right What happens with vulnerabilities found through Glasswing?
Responsible disclosure. Partners notify software maintainers first, allow time for patching, then publish details. The goal is to fix vulnerabilities, not create panic.
chevron_right Why not just open-source Mythos?
Because Mythos is as powerful offensively as it is defensively. Anthropic's assessment is that releasing the model without sufficient safeguards poses far more risk than benefit. Glasswing is the current compromise — let defenders use it first.
Stay Informed
Subscribe to the Claude Mythos intelligence network for fact-checked findings and critical updates.
No spam. Unsubscribe anytime.