Key Points:
- Anthropic has expanded its secure “Project Glasswing” initiative four-fold, extending advanced cyber-defense access to 150 additional organizations.
- The program now spans over 15 countries, bringing in critical security agencies like the EU’s ENISA, the Five Eyes alliance, and tech giants like Samsung.
- The participating partners have used the unreleased “Claude Mythos Preview” model to successfully identify more than 10,000 highly critical software flaws.
- To manage the massive computing costs of the model, which burned through $1 million in just three weeks, Anthropic has confidentially filed for an IPO.
The global battle over cybersecurity and artificial intelligence has taken a massive, defensive leap forward as the world’s leading model developers work to protect our shared digital infrastructure. On Tuesday, June 2, 2026, San Francisco-based AI pioneer Anthropic announced a major, four-fold expansion of its secure cyber-defense initiative, Project Glasswing. Under the updated rollout, the startup is extending access to its highly advanced, unreleased “Claude Mythos Preview” model to approximately 150 additional partner organizations across more than 15 countries. This dramatic expansion increases the total partner pool to roughly 200, representing a coordinated effort to secure the world’s most critical software from malicious exploitation before the underlying technology enters the public domain.
The urgency behind the expansion of Project Glasswing is directly tied to the staggering volume of undiscovered software vulnerabilities that current partners have unearthed. Since its initial launch in April 2026 with a tight cohort of about 50 founding partners, the unreleased Claude Mythos model has systematically scanned the world’s most critical software backbones. To date, these partners have utilized the model to successfully identify more than 10,000 security flaws classified as highly or critically severe across major operating systems and web browsers. Anthropic estimates that a single coordinated cyberattack exploiting these unpatched vulnerabilities before the launch of Project Glasswing could have impacted the daily lives of more than 100 million people globally.
To maximize the defensive reach of the program, the new expansion introduces an incredibly diverse, multinational base of strategic partners. The new list of 150 organizations includes critical defense agencies from the “Five Eyes” intelligence alliance, NATO member states, and major global hardware manufacturing conglomerates like Samsung Electronics and SK Hynix. Over the weekend, the European Union’s central cybersecurity agency, ENISA, also gained official access, making it the first EU body to join the elite defensive coalition. These major public and private organizations will integrate the unreleased model directly into their internal security pipelines, using its advanced capabilities to scan, patch, and harden both proprietary and open-source software infrastructures.
Understanding why Claude Mythos has caused such a massive stir across both the public and private sectors requires looking at its unprecedented, autonomous capabilities. Unlike traditional, rule-based scanning tools or simple AI coding assistants, Mythos represents a major breakthrough in automated vulnerability discovery and exploit generation. During internal red-teaming tests, the model successfully identified and exploited complex software flaws that had gone undetected by human experts for decades, including a 27-year-old system bug in OpenBSD. The AI can autonomously map out highly creative, multi-step cyberattack paths, prompting serious concerns that if an unshielded version of the model were to leak, hostile actors could easily turn it into a devastating cyber-weapon.
Operating a defensive model of this immense scale has also exposed the sheer, capital-intensive nature of advanced artificial intelligence. In a leaked internal briefing, Anthropic admitted that the Mythos model had burned through a staggering $1 million in raw computing and inference costs in just three weeks of active partner testing. To help its partners absorb these immense operational expenses, Anthropic originally committed up to $100 million in direct usage credits across the Project Glasswing ecosystem, alongside $4 million in cash donations to open-source security foundations. These spiraling computing costs highlight why the financial moat of the generative AI sector belongs strictly to a tiny handful of heavily funded, mega-cap companies.
The extreme capabilities of Claude Mythos have also triggered significant alarm bells at the highest levels of global finance and monetary policy. In April, immediately following the model’s initial preview, U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an urgent, highly confidential meeting with the chief executives of Wall Street’s largest banks. During the briefing, regulatory officials warned that the proliferation of such advanced, code-hacking models could pose a major threat to the global financial system if the technology fell into the hands of state-sponsored syndicates. This high-level intervention underscores why the government is maintaining such tight, monitored oversight over the Project Glasswing pipeline.
To fund this incredibly expensive development cycle and compete against rivals like OpenAI, Anthropic has recently taken a major step toward public capital markets. Financial sources confirm that the startup has confidentially filed for a potentially historic initial public offering (IPO) on the Nasdaq. Wall Street analysts project that the upcoming public listings of Anthropic, OpenAI, and SpaceX could collectively add up to a staggering $4 trillion in total U.S. stock market capitalization within the next few months. This massive capital influx is essential for these firms to secure the hundreds of thousands of advanced microchips and gigawatts of electricity required to train next-generation artificial intelligence models.
The launch of the extended Project Glasswing also represents a critical, system-level shift in how the global software industry manages security. Historically, software developers have relied on a “reactive” model of patching, where they only fixed security vulnerabilities after hackers had already exploited them in the wild. By providing a highly capable, autonomous AI to security defenders first, Anthropic is trying to establish a “proactive” paradigm of responsible disclosure. By using Mythos to find and patch thousands of high-severity vulnerabilities before the unshielded technology enters wide availability, the tech sector is actively building a highly resilient, modern digital defense.
Ultimately, the massive expansion of Project Glasswing to 150 additional partners marks a bold, highly strategic attempt to secure the world’s most critical software for the AI era. By giving elite defense agencies, global manufacturers, and international central banks direct access to the unreleased Claude Mythos model, Anthropic is proving that security and open collaboration must walk hand in hand. As the startup prepares to release “Mythos-class” models to all commercial customers in the coming weeks with additional, robust safety safeguards, the international community will watch closely. The final success of this historic project will define whether the tech community can successfully stay one step ahead of cyberattackers, or if the rapid advance of artificial intelligence will introduce a dangerous new era of digital instability.
