Anthropic was ordered by the US government to restrict access to its newest AI models, Fable 5 and Mythos 5, for all foreign nationals as well as domestic users and employees, invoking export control rules that industry experts find ambiguous and unprecedented in AI governance.
- Anthropic blocked AI model access globally under a US export control directive.
- The legal basis for the order is unclear, raising AI governance concerns.
- Experts call for clearer rules to define export controls on cloud-based AI.
What happened
Anthropic was compelled by an abrupt export control order from the Trump administration to restrict access to its latest AI models, Fable 5 and Mythos 5. The directive applied not only to foreign users but also impacted domestic users and Anthropic employees, forcing a complete shutdown of access to these models worldwide.
The government cited national security authorities as its basis for the order. However, Anthropic and external observers note that no clear transfer of physical model data or code occurs during regular AI usage, with users only interacting with the models via Anthropic’s servers remotely. This has created confusion about how the export controls are meant to be applied in this context.
Why it matters
This episode marks an unprecedented use of US export control laws applied to AI models accessed as cloud services rather than tangible goods or software exports. It exposes significant regulatory ambiguity around how evolving AI technologies fit within existing national security frameworks.
Experts warn that the lack of clarity places AI developers in a difficult position, unsure what safeguards are required or how their products might be restricted. It also contributes to international uncertainty about reliance on US-based AI platforms, potentially impacting global competitiveness and cooperation in emerging AI capabilities.
What to watch next
Stakeholders will be closely monitoring whether the US government provides more detailed guidance or legislation clarifying how export controls should apply to AI models and cloud-based services. Congressional actions are already underway to address known gaps in current export control regimes for digital technologies.
The AI industry will also watch for how major players like OpenAI, Google, and Meta respond to this precedent, as the ruling could influence restrictions on next-generation models. Ultimately, clearer policy is needed to balance national security concerns with the innovation and accessibility demands of frontier AI research.