AI safety and research company Anthropic has announced a significant expansion of Project Glasswing, its initiative to identify and resolve critical software vulnerabilities using artificial intelligence. The project is extending access to its powerful Claude Mythos model to approximately 150 new organizations across more than 15 countries. This move, which follows the company's recent confidential filing for an initial public offering, underscores a major push to bolster global cybersecurity defenses against emerging AI-driven threats.
A Proactive Stance on AI-Powered Cybersecurity
At the heart of Project Glasswing is Claude Mythos, an advanced AI model designed to scan codebases for security flaws with remarkable efficiency. During its initial phase with about 50 partners, including the U.S. government, the model successfully identified over 10,000 high or critical-severity vulnerabilities. This early success has validated the model's capabilities and provided the impetus for this broader rollout to secure the world's most important software.
The new cohort of partners was selected to cover industries that were not well-represented in the initial group, specifically those managing critical infrastructure. This includes organizations in the power, water, healthcare, communications, and hardware sectors, reflecting a strategic focus on protecting essential services. Many of these new partners are vendors whose codebases are foundational for countless other organizations and governments globally, amplifying the project's protective reach.
Global Collaboration for Critical Infrastructure
This expansion carries a significant international dimension, bringing in organizations from key U.S. allied nations to foster a united defensive front. Partners are based in countries including Australia, Canada, France, Germany, Japan, South Korea, and the United Kingdom, among others. By extending its geographical reach, Anthropic aims to create a global network dedicated to securing software that underpins both national and international security.
High-profile participants in the expanded project reportedly include identity management firm Okta, tech giants Samsung and SK Hynix, NATO, and the EU's cybersecurity agency, ENISA. Anthropic emphasizes that a successful cyberattack on any of these partners could have catastrophic consequences, potentially affecting more than 100 million people. The common thread among all participants is the critical nature of their software and its widespread impact on public safety.
Addressing an Evolving Threat Landscape
Anthropic's urgency is driven by the rapid advancement of AI capabilities across the industry, a trend the company has consistently highlighted. The firm anticipates that other AI developers will create models with Mythos-level cyber capabilities within the next six to twelve months. Project Glasswing is therefore a strategic effort to establish defensive norms and infrastructure before such powerful tools can be potentially misused by malicious actors.
The competitive environment is already active, with rivals like OpenAI also testing their own cybersecurity-focused models with partners. Anthropic's approach extends beyond just vulnerability detection, aiming to accelerate the entire security lifecycle, including patching and responsible disclosure. The company is also releasing some of its internal tools to trusted security teams to help the broader industry adapt to this new reality of AI-powered cyber defense.
The expansion of Project Glasswing marks a pivotal moment in the intersection of artificial intelligence and cybersecurity, shifting from theoretical potential to practical, large-scale application. By arming critical infrastructure providers and software maintainers with its most advanced AI, Anthropic is fostering a collaborative, global defense against increasingly sophisticated threats. This initiative represents a crucial step toward building a permanent advantage for defenders and reshaping security paradigms for the AI era.