June 3, 2026

‍

Anthropic has extended access to its Claude Mythos Preview model to approximately 150 new organizations across more than 15 countries, significantly expanding Project Glasswing, its controlled initiative to use Mythos for defensive cybersecurity. 

The announcement came Tuesday. The expansion is the first major update since the program's April 2026 launch.

The original April cohort included roughly 50 organizations. Founding partners included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and Anthropic itself.

Those partners have since used Mythos to scan codebases, and the results are significant:

• 10,000+ high- or critical-severity vulnerabilities identified
• $100M in Mythos usage credits committed by Anthropic
• $4M pledged to open-source security organizations

The June 2 expansion adds sectors largely absent from the first cohort: power, water, healthcare, communications, and hardware.

Why Anthropic is Keeping It Controlled

Mythos can identify zero-day vulnerabilities, software flaws unknown even to the developers who built them, at machine speed. That capability is exactly why Anthropic isn't releasing it openly.

The company put the stakes plainly: "A successful attack on their codebase could be catastrophic. For most partners, we estimate that a major attack could affect more than 100 million people, with important ramifications for both global and national security."

The same capabilities that find vulnerabilities defensively could, in the wrong hands, be used to exploit them.

The controlled rollout has drawn criticism from prominent voices in AI safety and international policy.

Yoshua Bengio, Turing Award laureate and chair of the 2026 International AI Safety Report, told Fortune:

"It doesn't make sense that private individuals are deciding the fate of infrastructure for everyone else. What about all the companies and all the countries that didn't get access?"

Bengio called for a regulatory body comparable to the FDA to oversee frontier AI development. "There's no reason that it's going to limit itself to attacking U.S. infrastructure or U.S. citizens," he said. "So this has to be an international affair."

Internationally, the pressure is mounting:

• The Bank of England publicly pressed Anthropic for access on behalf of U.K. banks
• IMF and World Bank spring meeting discussions were unexpectedly dominated by Mythos concerns
• European regulators warned they had yet to gain access or understand the scale of vulnerabilities uncovered
• A White House memo instructed Defense, Treasury, and Homeland Security to prepare for federal access

What Mythos is Actually Being Used For

Initial partners aren't just running vulnerability scans. Glasswing organizations are deploying Mythos for:

• Automated patch writing
• Penetration testing simulations
• Pre-release security checks
• Threat detection automation
• Rebuilding legacy codebases in memory-safe languages

The Cloud Security Alliance, in a briefing led by Gadi Evron of Knostic with contributions from former CISA Director Jen Easterly, Bruce Schneier, and former National Cyber Director Chris Inglis, issued a stark warning:

"In the near term, security organizations will likely be overwhelmed by the need to apply patches and respond to AI-discovered vulnerabilities, exploits, and autonomous attacks. The storm of vulnerability disclosures from Project Glasswing is the first of many large waves."

Anthropic confirmed it is planning a broader release of Mythos, stating that "hundreds of thousands of organizations, researchers, and maintainers will likely need access" to defend against adversaries operating comparable models.

"Project Glasswing has taught us a great deal about how to respond when models cross important capability thresholds. If we're successful, we hope to enable a permanent advantage for defenders."

Bob Zukis (Digital Directors Network) and Jesse H. Webb, CISO at Avalon Healthcare Solutions, wrote in a Harvard Law School briefing that boards should "require management to translate vendor capability into enterprise defense" and distinguish real shifts from noise.

Enterprise teams are already reviewing AI governance and access control frameworks in response, with particular focus on visibility into internal AI tool usage. CloudEagle.ai, which provides shadow AI detection, AI tool blocking, and token consumption tracking, is among the platforms seeing increased interest from security and IT teams looking to close the governance gap.

‍

Shadow AI Isn't Free

CloudEagle.ai reveals who's using AI and what it's costing.

See How CloudEagle Works

Book a Demo