The European Union is seeking to intensify discussions with the United States over access to the world's most advanced artificial intelligence models, particularly systems with powerful cybersecurity capabilities, as policymakers seek greater oversight of rapidly advancing technology.
The push comes after Anthropic's Mythos model, unveiled in April, sparked significant concern among governments and businesses due to its potential cyber capabilities and the possibility that such systems could be used to identify vulnerabilities or facilitate sophisticated cyberattacks.
Anthropic said on Thursday it expects to make "Mythos-class" models available to customers in the coming weeks, marking a significant step in the broader rollout of the technology.
The White House has been working closely with major AI developers as it seeks to balance technological innovation with national security considerations, while maintaining the United States' lead over China in the global artificial intelligence race.
Anthropic initially provided preview access to Mythos through a programme known as Project Glasswing, limiting availability to a select group of companies and organisations.
However, the company has not granted preview access to the European Union, its AI Office or government agencies outside the United States, with the exception of the United Kingdom's AI Security Institute.
According to previous reports, Anthropic had proposed expanding access to approximately 70 additional organisations, although administration officials reportedly raised concerns about the security implications of broader distribution.
"Cybersecurity is a shared priority and we have agreed to mutually recognise our respective standards in this area," Thomas Regnier, Commission spokesperson, told CNBC, referring to discussions between the European Commission and the U.S. administration.
"In parallel we are now expanding technical discussions with Anthropic and other model developers which have notified their latest models to the AI office."
An anonymous source also told CNBC that the AI company indicated that the bloc would need approval from the U.S. administration before gaining preview access to Mythos.
The source, who spoke on condition of anonymity because the discussions are confidential, said Anthropic had suggested the White House was not specifically opposed to sharing Mythos with the European Union, but was more broadly reluctant to provide access to non-U.S. government entities.
Anthropic said that models with capabilities comparable to Mythos require robust cybersecurity safeguards before they can be released broadly.
"We're making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks," the company said.
Earlier this month, Anthropic Chief Executive Officer Dario Amodei warned that developers may have only a limited window to address vulnerabilities identified by advanced AI systems before rival models, particularly from China, achieve similar capabilities.
Amodei said there was a six-to-12-month period to patch tens of thousands of software vulnerabilities identified by Mythos before competing Chinese AI models potentially reach comparable levels of performance.
