Anthropic Expands Claude AI Access to 150 New Organizations

Key Takeaways

• Anthropic has significantly expanded its AI-driven cybersecurity initiative, Project Glasswing, adding approximately 150 new organizations to its Claude Mythos Preview program.
• The expansion includes critical infrastructure providers, hardware manufacturers, and organizations maintaining widely used codebases across more than fifteen countries.
• Early partners identified over 10,000 high- or critical-severity security flaws, demonstrating the AI’s robust vulnerability detection capabilities.
• Anthropic warns that similar “Mythos-class” AI models from competitors could emerge within six to twelve months, potentially without adequate safeguards against offensive misuse.
• The company is also rolling out broader defensive tools and a public product, Claude Security, while working towards a general access release with robust safeguards.

Anthropic Broadens Claude AI Cybersecurity Initiative, Onboards 150 New Organizations

Anthropic has substantially broadened the scope of Project Glasswing, its collaborative AI-powered cybersecurity endeavor, by integrating approximately 150 additional organizations into its Claude Mythos Preview program. This expansion follows an intensive period of collaboration with existing partners, key players in the security industry, maintainers of open-source software, and the U.S. government.

Project Glasswing initially launched in early April 2026, granting roughly 50 founding partner organizations access to Claude Mythos Preview. These initial participants utilized the AI to scan their codebases for potential security vulnerabilities. The efficacy of this early phase was clear, with these partners collectively uncovering more than 10,000 security flaws deemed high or critical in severity. This outcome powerfully underscores the advanced detection capabilities of Mythos-class AI models when applied to complex enterprise and infrastructure codebases.

The newly admitted organizations hail from over fifteen countries, with a significant portion delivering vital critical infrastructure services that impact global operations. The expansion strategically includes sectors previously underrepresented in the initial cohort, such as power, water, healthcare, communications, and hardware industries. Furthermore, a considerable number of these new partners are vendors and non-profit entities responsible for maintaining widely adopted codebases, which are relied upon by numerous other organizations worldwide, including national governments.

Anthropic’s assessment of the inherent risks is stark: for many Project Glasswing partners, a successful cyberattack targeting their codebases could potentially impact over 100 million individuals, leading to severe ramifications for both national and global security. Consequently, each new organization must rigorously satisfy Anthropic’s stringent security protocols before being granted access to these advanced AI models.

Claude Mythos Preview’s Evolving Capabilities

Beyond its core function of raw vulnerability detection, partners are now leveraging Claude Mythos Preview for a more extensive array of defensive tasks. These applications include automated patch generation, pre-release security evaluations to prevent vulnerabilities from reaching production environments, sophisticated penetration testing simulations, automation of threat detection and response mechanisms, and the refactoring of legacy codebases into memory-safe languages.

Mythos Preview also serves as a critical harbinger for the cybersecurity industry. Anthropic projects that within the next six to twelve months, rival AI companies will likely develop their own Mythos-class models. There is concern that these models might be released without adequate safeguards to prevent their misuse for offensive cyber operations. This impending timeline creates a narrow, urgent window for cyber defenders to adapt their tools and operational strategies before the threat landscape undergoes a dramatic transformation.

To disseminate these defensive capabilities beyond the direct participants of Project Glasswing, Anthropic is making the tools developed to support the program’s vulnerability detection workflows available upon request to trusted security teams. Additionally, the company has introduced Claude Security, a new product built upon its frontier public models, including Claude Opus 4.8. This offering is specifically designed to enable organizations outside the direct program to scan their codebases and receive suggestions for patches.

Anthropic is also engaged in active discussions with various third parties to significantly scale up the review and patching of vulnerabilities within open-source software. This includes efforts to establish standardized best practices for disclosing findings to open-source maintainers, thereby streamlining the triage and remediation processes. The company acknowledges that hundreds of thousands of organizations will ultimately require access to Mythos-level capabilities to effectively address the full scope of the evolving threat environment.

While Anthropic is working towards a general access release, it emphasizes that highly robust safeguards to prevent offensive misuse—safeguards that neither Anthropic nor other AI developers have fully perfected yet—must first be firmly in place. Concurrently, Anthropic plans to further expand its Cyber Verification Program, which will grant Mythos-class capabilities to additional organizations for specific defensive tasks. The overarching objective of Project Glasswing remains consistent: to evolve AI’s role in cybersecurity from merely identifying vulnerabilities to a comprehensive, full-cycle model encompassing disclosure, patching, and deployment, ultimately securing a lasting advantage for cyber defenders.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.