Anthropic is expanding access to its powerful Claude Mythos AI model, inviting approximately 150 more organizations to join Project Glasswing. The AI company announced the expansion today, bringing partners from more than 15 countries into the exclusive testing program.
The expansion comes two months after Anthropic first introduced Claude Mythos and Project Glasswing in April. Unlike typical AI model releases, Anthropic kept Mythos restricted to a select group of organizations due to its advanced cybersecurity capabilities. The company explained that AI models have reached a point where they can outperform most humans at finding and exploiting software vulnerabilities.
The new batch of partners represents a shift from the original cohort, which included tech giants like Apple, Amazon, Microsoft, and NVIDIA. This time, Anthropic is focusing on critical infrastructure providers that weren’t well-represented in the initial group:
- Public utilities managing power, water, and telecommunications
- Healthcare providers and medical institutions
- Other organizations whose systems could affect millions if compromised
Each new partner must pass Anthropic’s security requirements before gaining access to Claude Mythos. The company notes that what these organizations have in common is that a successful cyberattack on their systems could be catastrophic, potentially affecting more than 100 million people and carrying serious implications for national and global security.
This careful, controlled rollout reflects growing concerns about AI’s potential for both defensive and offensive cybersecurity applications. While Claude Mythos can help organizations identify and patch vulnerabilities in their code, the same capabilities could theoretically be misused by bad actors to find exploits.
Anthropic says it’s working to bring Mythos-level capabilities to the public but won’t make the model widely available until it implements robust safeguards to prevent misuse. The company is taking a cautious approach that mirrors broader industry debates about releasing powerful AI models with dual-use potential.
The announcement of Claude Mythos has already had political ripple effects. In May, reports suggested that the model’s cybersecurity capabilities prompted the Trump administration to reconsider its stance on AI regulation. The president was reportedly set to sign an executive order creating a federal process for sharing AI-discovered software vulnerabilities with critical institutions before public model releases. However, the signing was postponed at the last minute due to objections from Trump and the tech industry.
This regulatory uncertainty highlights the complex balance companies like Anthropic must strike. They need to demonstrate their AI systems’ beneficial uses while addressing legitimate concerns about weaponization. Project Glasswing represents one approach: giving critical infrastructure operators early access to defensive capabilities while keeping potentially dangerous tools from spreading too quickly.
