Anthropic has expanded access to its next-generation artificial intelligence model, Claude Mytos, to approximately 150 companies and institutions across 15 countries as part of a broader effort to strengthen cybersecurity preparedness and evaluate the capabilities of advanced AI systems.
The company announced on June 2 (local time) that it is expanding its Project Glasswing partnership following several weeks of close collaboration with the U.S. government and industry partners.
Project Glasswing grants participating organizations access to a preview version of Claude Mytos, allowing them to evaluate how next-generation AI models may be deployed, identify potential misuse scenarios, and assess cybersecurity vulnerabilities in widely used software and critical infrastructure systems. The initiative was initially limited to around 50 organizations, including U.S. government agencies and the UK's AI Security Institute.
The newly added partners span a broad range of sectors, including hardware manufacturing, energy, water utilities, healthcare, and telecommunications. Organizations must meet strict security requirements before receiving access to the model.
Anthropic said that while the new participants are based in 15 countries, many operate critical systems whose disruption could affect more than 100 million people and have implications for both national and global security.
The company said the expansion reflects its expectation that other AI developers will release models with capabilities comparable to Mytos within the next six to twelve months. Through the preview program, Anthropic aims to help organizations prepare for a future in which advanced AI systems become widely available, including scenarios where such models may be used without built-in safeguards against misuse.
A major focus of the initiative is preparing for increasingly sophisticated and unpredictable cyberattacks. Participating organizations are expected to develop both deployment strategies and defensive measures for the next generation of AI-powered threats.
Anthropic also plans to provide secure access to advanced models, tools, and shared infrastructure to help the software industry adapt to rapidly evolving AI capabilities. The company will support the disclosure, remediation, and distribution of software patches when vulnerabilities are identified.
According to Anthropic, participants in the program have discovered approximately 10,000 software vulnerabilities worldwide over the past month, with 66% classified as critical security threats.
The company did not disclose the identities of newly participating countries or organizations. However, according to reports from international media outlets including the Financial Times, South Korea has joined the initiative alongside several European countries, including France, Germany, Italy, the Netherlands, Belgium, and Sweden.
In South Korea, the Korea Internet & Security Agency (KISA) is reportedly participating on behalf of the government, alongside major corporations such as Samsung Electronics, SK Hynix, and SK Telecom.
