Anthropic has been forced to shut down access to its latest AI models for foreign nationals, including those within the US. The Trump administration’s opaque export controls have left the company scrambling to comply with an order it can barely understand.

The move, which has no precedent in AI governance, highlights the challenges of regulating something as intangible and global as artificial intelligence. While traditional export controls apply to discrete items like hardware or software code, AI models are more akin to services that can be accessed remotely. This leaves companies unsure how to navigate the complex regulatory landscape.

Experts warn that such ad hoc interventions are unsustainable and could jeopardize America’s global lead in AI development. The incident raises questions about whether Anthropic was targeted due to unique capabilities, specific safeguard issues or its political relationships with the Trump administration. Regardless of the reasoning, it underscores a need for clearer guidelines from policymakers.

The episode also fuels concerns among governments and companies outside the US that they should not rely on American firms for strategic systems. If creating models that can’t be jailbroken becomes a de facto standard in the US, American developers may find themselves at a disadvantage.