The European Union is seeking to deepen discussions with Anthropic and the U.S. administration regarding access to some of the world’s most advanced AI models, including systems with significant cybersecurity capabilities.
At the center of the discussions is Anthropic’s Mythos model, unveiled earlier this year, which has attracted attention from governments and enterprises due to its ability to identify software vulnerabilities and support advanced cybersecurity analysis.
According to EU officials, the bloc is expanding technical conversations with Anthropic and other frontier model developers as regulators seek greater visibility into cutting-edge AI systems that could have significant economic and security implications.
Anthropic initially limited Mythos access through its Project Glasswing initiative, providing preview access to a small group of organizations. The company has not yet extended review access to the EU’s AI Office or most non-U.S. government entities, with the exception of the U.K.’s AI Security Institute.
The discussions come amid broader debates over balancing AI innovation, national security, and international oversight. U.S. officials have emphasized the importance of maintaining America’s leadership in artificial intelligence while carefully managing the risks associated with highly capable frontier models.
Anthropic CEO Dario Amodei recently warned that advanced AI systems are rapidly improving cybersecurity capabilities and could expose large numbers of software vulnerabilities, increasing urgency around both defensive measures and governance frameworks.
The talks also reflect Anthropic’s growing influence in the global AI landscape. The company recently released Claude Opus 4.8, introducing improved performance, reliability, and enhanced developer controls, while a $65 billion Series H funding round valued the company at $965 billion, surpassing OpenAI and making Anthropic Silicon Valley’s most valuable AI company. As Anthropic’s frontier models become increasingly powerful, governments and regulators are seeking a greater role in understanding how these systems are deployed and governed.
