Anthropic Expands Glasswing AI Cybersecurity to 150 New Partners
Anthropic is significantly scaling up its AI-powered cybersecurity initiative, Project Glasswing, by onboarding approximately 150 new organizations across more than 15 countries. These entities include providers of critical infrastructure in sectors like power, water, healthcare, and communications, marking a major stride in the fight against escalating cyber threats. The announcement, made on June 2, follows a successful pilot program involving 50 partners that uncovered over 10,000 high-severity vulnerabilities using Anthropic's Claude Mythos Preview model.
Initially launched in April 2026, Project Glasswing leverages advanced AI to proactively identify and remediate software vulnerabilities before they can be weaponized. Early participants included tech giants such as Microsoft, Google, and Cisco, alongside critical infrastructure players like Broadcom and the Linux Foundation. The addition of this new cohort expands the project’s global footprint and industry coverage, introducing key players in hardware manufacturing and open-source software maintenance to the initiative.
The stakes are high. Anthropic estimates that a successful attack on the codebases of participating organizations could impact over 100 million people globally, with implications for both national and international security. As cheap, powerful AI models with cyber capabilities loom on the horizon, this effort is as much about defense as it is about preparing for a rapidly shifting threat environment. "Within 6 to 12 months, Mythos-class models may proliferate without the necessary safeguards," Anthropic warned, emphasizing the urgency of scaling up defensive measures.
Claude Mythos Preview: AI at the Forefront of Cyber Defense
At the heart of Project Glasswing is Claude Mythos Preview, a cutting-edge AI model designed to identify critical vulnerabilities and suggest fixes. Unlike traditional static analysis tools, Mythos Preview excels in penetration testing and detecting complex exploit paths, making it highly effective for securing modern codebases. Anthropic has also introduced Claude Security, a companion product built on Claude Opus 4.8, to further streamline the patching process.
Participants in the initial phase of Glasswing not only scanned their systems for vulnerabilities but also began using Mythos Preview to write patches and conduct pre-release security checks. This AI-driven approach is now being scaled to the new partners, many of whom maintain software relied upon by governments and millions of end-users worldwide.
Cybersecurity at a Crossroads
Anthropic’s expansion of Project Glasswing comes at a critical time. The cybersecurity landscape is evolving rapidly, with AI tools enabling both defenders and attackers. Glasswing aims to tip the scales in favor of the defenders by accelerating vulnerability detection and patching, while also fostering global collaboration among security teams, researchers, and open-source maintainers.
One bottleneck the initiative seeks to address is the lag between discovering vulnerabilities and deploying fixes—a delay that malicious actors could exploit. To tackle this, Anthropic is working with third parties to scale vulnerability review and patching processes, particularly for open-source software. By sharing best practices and providing advanced tools, the company aims to establish new industry standards for cybersecurity in the AI era.
Looking Ahead
Anthropic’s ambitions don’t stop here. Future expansions of Project Glasswing will prioritize additional critical infrastructure providers and open-source maintainers, with plans to broaden its geographical reach. The company is also actively developing robust safeguards to ensure that AI models like Mythos Preview are not misused—a challenge Anthropic acknowledges has yet to be fully solved by the AI industry.
As AI continues to redefine the cybersecurity landscape, Anthropic is positioning Project Glasswing as a cornerstone for global cyber defense. If successful, the initiative could provide a lasting advantage for defenders, setting new benchmarks in securing the world’s most essential software systems.