The US government lifted export controls on Anthropic's AI models this week. But the most advanced model, Mythos 5, remains limited to a few vetted institutions such as the UK's AI Security Institute and Japanese banks, while European banks are still denied access.
"The US is essentially acting as a gatekeeper for the most advanced AI," a Commerce Department spokesperson said. The move highlights how much US allies' digital security depends on American policy.
Mythos 5 is extremely powerful. When Anthropic first previewed it in April, the company limited access to only a small group of trusted institutions for cybersecurity testing. That was a voluntary restriction by Anthropic, a private company.
Get your free investing masterclass bonus when you join Market Briefs, our free daily newsletter
Then the US government added its own export controls on top. On June 12, the US Commerce Department put in place export controls that forced Anthropic to seek authorization before any non-US citizen could use Fable or Mythos. Washington's policy shows how much US allies' digital security depends on American decisions.
Through a program called Project Glasswing, Anthropic added organizations providing critical infrastructure from more than 15 countries by early June. Yet those organizations are mostly domestic or carefully vetted.
That means even close allies in Europe are left out. The gap between who gets Mythos and who doesn't is wide. The Trump administration kept the restrictions in place, which meant foreign organizations remained unable to obtain access.
The restriction on Mythos 5 underscores the growing concern among US policymakers about the dual-use nature of frontier AI models. By limiting access to trusted partners, the US aims to prevent adversaries from exploiting the technology while still advancing AI safety collaboration with close allies. The UK's AI Security Institute and Japanese banks have been granted access, but European financial institutions remain excluded, highlighting a divide that may affect international cooperation on AI governance.
Project Glasswing, launched earlier this year, was designed to grant select international partners early access to Anthropic's frontier models for cybersecurity testing. However, the initiative has been criticized for its limited scope, with only 15 countries' critical infrastructure organizations included by June. The exclusion of European banks, despite their close alignment with US security interests, has raised questions about the criteria for trustworthiness and the potential for creating a two-tiered system among allies.
Subscribe to Market Briefs, our free daily newsletter, and claim your bonus investing masterclass
