Anthropic Launches Project Glasswing: 12 Tech Giants Using AI to Secure Critical Software

Anthropic Project Glasswing cybersecurity coalition of 12 tech companies using AI to find zero-day vulnerabilities

Anthropic has launched Project Glasswing, a major cybersecurity initiative bringing together 12 of the world's most influential technology companies to use AI offensively — in defense. The coalition includes Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, along with more than 40 additional organizations managing critical software systems. The project centers on Claude Mythos Preview, an unreleased frontier AI model that Anthropic says "can surpass all but the most skilled humans at finding and exploiting software vulnerabilities." The initiative represents the most significant coordinated industry response to date to the threat of AI-enabled cyberattacks — and the clearest signal yet that the cybersecurity implications of Claude Mythos are being taken seriously by the industry itself.

What Claude Mythos Has Already Found

Project Glasswing is not theoretical. Before the initiative was publicly announced, Claude Mythos Preview was already deployed against production codebases — and the results are striking. The model identified thousands of zero-day vulnerabilities across major operating systems, web browsers, and critical infrastructure components, including previously unknown flaws in OpenBSD (a 27-year-old codebase), FFmpeg (16 years old), and the Linux kernel. On cybersecurity vulnerability reproduction benchmarks, Mythos Preview achieves an 83.1% success rate — significantly outperforming previous versions and, according to Anthropic, most human security researchers.

The selection of targets is deliberate. OpenBSD and the Linux kernel underpin enormous swaths of the world's server infrastructure. FFmpeg processes video across billions of devices. Finding exploitable vulnerabilities in these codebases — and disclosing them to maintainers before attackers can use them — is precisely the kind of high-leverage defensive action that Project Glasswing is designed to systematize. As Microsoft's own AI-enhanced security response center has shown, AI tools applied to vulnerability discovery can dramatically compress the timeline between flaw introduction and patch deployment.

The Strategic Logic Behind the Coalition

Anthropic is backing the initiative with $100 million in Claude model usage credits distributed to participating organizations, plus $4 million in direct donations to open-source security programs through the Linux Foundation and Apache Software Foundation. The financial commitment signals that Project Glasswing is structured as a long-term program rather than a one-time disclosure exercise.

The coalition's composition is also significant. By bringing together hardware makers (NVIDIA, Broadcom), cloud providers (AWS, Google, Microsoft), security specialists (CrowdStrike, Palo Alto Networks), enterprise software users (JPMorganChase), and open-source stewards (Linux Foundation), Anthropic has created a coalition that spans the full software supply chain. The logic is that a vulnerability in a foundational library like FFmpeg affects every layer of the stack simultaneously — and the fix needs buy-in from every layer to propagate effectively. This connects directly to growing concern about government-level AI security risks that prompted a White House emergency meeting with tech CEOs ahead of the Mythos release.

Frequently Asked Questions

What is Project Glasswing?

Project Glasswing is a cybersecurity coalition launched by Anthropic with 12 founding technology and finance companies. It uses Claude Mythos Preview AI to proactively identify zero-day vulnerabilities in critical software and disclose them to maintainers before attackers can exploit them.

Which companies are involved in Project Glasswing?

The 12 founding partners are Anthropic, AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, alongside more than 40 additional participating organizations.

What is Claude Mythos Preview?

Claude Mythos Preview is an unreleased frontier AI model from Anthropic that demonstrates exceptional capability in identifying and exploiting software vulnerabilities. It scores 83.1% on cybersecurity vulnerability reproduction benchmarks and has already found thousands of zero-day flaws in widely used open-source software.

The Bottom Line

Project Glasswing is Anthropic's answer to a question the AI industry has been reluctant to address head-on: if AI models can find and exploit software vulnerabilities faster than humans, who is going to use them first? By forming a coalition of 12 major technology companies and deploying Mythos Preview's capabilities defensively — finding flaws before attackers do — Anthropic is making a structural bet that the right response to AI-enabled offense is AI-enabled defense at scale. With $100 million in model credits committed and zero-days already found in the Linux kernel and OpenBSD, Project Glasswing is not a press release. It is a new category of cybersecurity infrastructure.