Anthropic disabled foreign national access to its Fable 5 and Mythos 5 models on June 12, 2026, after the US government issued an export control directive requiring the suspension, according to Anthropic's official notice and reporting by CNBC.

The directive is specifically scoped to foreign nationals — not geographic regions or foreign-incorporated entities as such, but individuals without US nationality, wherever they may be operating. That distinction matters. Export controls in the EAR (Export Administration Regulations) framework have historically targeted technology transfer by destination country, but a nationality-based restriction is a sharper instrument: it follows the person, not the IP address. Al Jazeera confirmed the foreign-nationals framing, and nothing in the sourcing suggests country-specific carve-outs.

Fable 5 and Mythos 5 are Anthropic's frontier model offerings. A directive of this scope — covering two named production models simultaneously — implies a formal Commerce Department or executive-branch determination that these systems meet the threshold for controlled technology under US law, likely classified under the dual-use framework that has governed advanced semiconductor exports since the October 2022 and October 2023 BIS rules.

The mechanics of compliance will be non-trivial for Anthropic. Nationality-based access controls are harder to enforce than geographic IP blocks; they require affirmative verification of user nationality at the account level, which most consumer and API platforms are not built to perform at scale. Whether Anthropic will implement a nationality attestation workflow, suspend all non-verified accounts by default, or simply gate new sign-ups while grandfathering existing users is not specified in the sourcing available.

For enterprise customers — particularly multinational firms that route AI workloads through centralized API keys shared across international teams — the operational disruption could be immediate. A foreign national employee accessing Fable 5 or Mythos 5 through a US parent company's account would, under a strict reading of the directive, be a compliance violation. Legal and procurement teams at affected organizations will need to assess whether their current usage patterns satisfy the directive's requirements, with some urgency.

The broader policy context is one of accelerating AI-specific export control activity. The Biden administration moved to layer AI model weights and API access into the export control architecture in its final months; the current administration has retained and, by the evidence of this directive, extended that posture. Restricting access to named frontier models by nationality is a step beyond chip export controls — it targets the inference layer directly, not just the hardware needed to train or run the models domestically. That is a meaningful escalation in the regulatory architecture governing advanced AI.

What this means for Anthropic's international revenue and competitive positioning against non-US frontier model providers — who face no equivalent domestic restriction — is a legitimate open question. The directive imposes asymmetric compliance costs on US-based AI labs relative to their Chinese, European, and open-weight competitors. Whether that trade-off reflects a deliberate strategic calculus by the administration, or is an artifact of applying a Cold War-era export control framework to a technology it was not designed to govern, will likely be the central debate as this directive gets scrutinized by Congress, affected businesses, and trading partners.

For now, Anthropic is in compliance mode. The models are off. The policy is live.