The Trump administration appeared ready to allow Anthropic to restore access to its Fable 5 model as of late June 2026, ending a two-week freeze. The freeze had been triggered by a government directive blocking foreign nationals from using Fable 5 and Mythos 5, Reuters reported.

The restriction started when the Trump administration issued a directive in mid-June requiring Anthropic to block foreign national access to both models, according to AP News. On the same day, Anthropic said it had begun working with the US government to restore access. For any company running these models across borders — whether serving international clients or coordinating research teams spread across multiple countries — the sudden cutoff created immediate operational headaches.

The government's approach to unfreezing access was uneven. On June 26, the US government sent a letter to Anthropic co-founder Tom Brown approving restored access to Mythos 5, but made no such approval for Fable 5, CNBC reported. That left the two models on different approval tracks, with Fable 5 still awaiting clearance a day later.

How Anthropic actually enforces nationality-based access controls is not clear from public statements, but the requirement itself points to the API layer — the interface through which customers request the model's output. Typically, such restrictions live in customer contracts rather than being enforced in real time at the moment of use. Building that kind of on-the-fly nationality verification is a substantial engineering and compliance challenge. Whether the government specified exactly how Anthropic should do this, or left that decision to the company, has not been disclosed.

The larger picture here involves a shift in how the US government thinks about cutting-edge AI models. Increasingly, Washington treats them as export-sensitive technology — similar to how it regulates advanced semiconductors or military software. Existing export control frameworks (ITAR and EAR, which stand for International Traffic in Arms Regulations and Export Administration Regulations) were built for hardware and software with obvious military uses. Applying those same rules to general-purpose language models creates real gray areas: the same tool that can speed up drug discovery could theoretically help with tasks the government wants to restrict.

The fact that Mythos 5 and Fable 5 received different treatment suggests the government is evaluating each model's capabilities separately, rather than applying a blanket rule to all frontier-tier models. If that becomes standard practice, every new model release at the cutting edge could require its own government review and approval process. That would add a new layer of compliance burden for Anthropic and any other AI company releasing state-of-the-art models.

For companies using these models in production — especially those with teams or customers overseas — the two-week freeze highlights a real business risk. Having restricted access to a core tool with no clear public timeline for restoration is the kind of supply-chain disruption that teams responsible for architecture and purchasing now need to plan for. Strategies for building redundancy and avoiding dependence on a single model started to feel much more practical after this incident.

Anthropic has framed itself as a willing partner with the government on these restrictions, consistent with its public stance on AI safety and its relationships with federal agencies. Whether that cooperative approach helped speed the approval process, or whether the timeline simply reflected how quickly the government could review the request, is impossible to say from public information alone.