Project Glasswing

source Apr 8, 2026

cybersecurityanthropicai-safetycoalitionopen-source

Anthropic’s announcement of Project Glasswing, a cross-industry initiative to apply frontier AI capabilities to defensive cybersecurity.

Coalition

Twelve launch partners: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Over 40 additional organizations received access to scan critical software infrastructure.

Central claim

AI models have reached a capability level where they surpass all but the most skilled humans at finding and exploiting software vulnerabilities. Claude Mythos Preview demonstrates this with thousands of zero-day discoveries across every major OS and browser. The capability will proliferate — Project Glasswing aims to give defenders a head start.

Key evidence

27-year-old vulnerability in OpenBSD (remote crash via network connection)
16-year-old FFmpeg vulnerability in a line hit 5 million times by automated testing
Autonomous discovery and chaining of multiple Linux kernel vulnerabilities for privilege escalation
All found without human steering

Commitments

$100M in model usage credits for participants
$2.5M to Alpha-Omega and OpenSSF (via Linux Foundation)
$1.5M to the Apache Software Foundation
Public report on findings within 90 days
Recommendations covering vulnerability disclosure, patching automation, supply-chain security, secure-by-design practices

Access model

Mythos Preview is not planned for general availability. After the research preview, pricing is $25/$125 per million input/output tokens. Available via Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. A Cyber Verification Program will give security professionals access through future safeguarded models.

Implications

The article frames AI-assisted vulnerability discovery as a dual-use capability: dangerous in adversary hands, invaluable for defense. The defensive AI advantage thesis holds only if defenders act before capabilities proliferate. Anthropic positions this as a national security priority for democratic states and calls for an independent third-party body to coordinate long-term cybersecurity efforts.

AI Cyber Proliferation Claude Mythos Preview represents a threshold: a model that surpasses all but the most skilled humans. Anthropic chose not to release it generally and instead channeled it into Project Glasswing. But the announcement itself signals that this capability class exists and will recur in future models from multiple developers. →

AI Cyber Proliferation The Glasswing announcement estimates global cybercrime costs at roughly $500B per year under current conditions. State-sponsored actors (China, Iran, North Korea, Russia) already target critical infrastructure. AI-augmented attacks could make these more frequent, more sophisticated, and accessible to less-skilled actors. →

AI Cyber Proliferation

Temporal advantage: deploy defensive scanning before adversaries gain equivalent capabilities (defensive AI advantage)
Access control: restrict the most capable models (Anthropic's approach with Mythos Preview)
Safeguard development: detect and block dangerous outputs before making Mythos-class models broadly available
Coalition defense: coordinate scanning across the full attack surface (Project Glasswing)
Government involvement: maintain democratic nations' lead in AI technology; assess and mitigate national security risks

→

AI Cyber Proliferation

Defensive AI Advantage — the counter-strategy
AI Vulnerability Discovery — the capability that proliferates
Project Glasswing — the coalition response

→

AI Vulnerability Discovery Claude Mythos Preview demonstrated this capability at scale during Project Glasswing: →

Claude Mythos Preview Claude Mythos Preview is an unreleased frontier model trained by Anthropic, announced as part of Project Glasswing. Named from Ancient Greek for "utterance" or "narrative." Not planned for general availability. →

Claude Mythos Preview

Project Glasswing — the initiative built around this model
AI Vulnerability Discovery — the capability it demonstrates

→

Defensive AI Advantage

Speed. Defenders must deploy AI-assisted scanning before capabilities proliferate to adversaries.
Coordination. No single organization can cover the full attack surface — coalition efforts like Project Glasswing are necessary.
Open-source coverage. Most critical infrastructure runs on open-source software maintained by under-resourced teams. The defensive advantage collapses if these codebases are left unscanned.

→