‹ rbtfl.
⋯
LeadersConflictsDefenceAIEnergySpaceTradeMineralsShippingMoneyDebtFood all trackers ▸
US bars Anthropic's most capable models from every foreign national worldwide

US bars Anthropic's most capable models from every foreign national worldwide

Fable 5 and Mythos 5, pulled three days after launch, set the first precedent for nationalising AI model access; Congress demands written justification by June 26

AI·Courts· active Who Decides·The Long Game ·8 takes · ·rbtfl upd Jun 25, 2026

Summary

On 12 June 2026, three days after launch, Anthropic suspended all access to Claude Fable 5 and its underlying model Mythos 5 under a binding US government export-control directive. The order bars any foreign national, anywhere on the planet, including non-citizen Anthropic employees, from accessing the models. Anthropic, unable to filter users by nationality in real time across dozens of cloud platforms, shut the models for all users globally. The government rationale: NSA Director Gen. Joshua Rudd told a Senate Intelligence Committee that Mythos identified vulnerabilities in "nearly all NSA classified systems within hours" after an adversary used a jailbreak. Anthropic publicly disagreed, characterising the jailbreak as "a narrow potential" and calling the recall of a model serving hundreds of millions unprecedented. Four bipartisan Congress members demanded written Commerce Department justification by 26 June. Anthropic's earlier Claude Opus 4.8 remains live.

The split

Washington's national-security framing holds that a model capable of autonomous cyberweapon development cannot be treated as ordinary software, the Export Administration Regulations' "specially designed" clause is being stretched to cover API-delivered AI capabilities for the first time. Anthropic frames the directive as a category error: the cited jailbreak involves asking the model to review and fix code, a function marketed as a feature. Google and OpenAI have not commented publicly, but the legal precedent, if upheld, means every frontier lab is a potential target of the same authority. Al Jazeera and non-US commentary note the geopolitical asymmetry: Deepseek and Alibaba Qwen face no reciprocal US pressure, while the directive effectively segments the world's most capable AI along national lines, accelerating the bifurcation the G7 Évian summit tried to manage through its "trusted partners" scheme.

By the numbers

  • June 9, Fable 5 and Mythos 5 launched.
  • June 12, 5:21 PM ET, suspension takes effect.
  • 3 days, time between launch and global recall.
  • $965B, Anthropic valuation at IPO filing.
  • 100M+, users globally affected by the suspension.
  • June 26, deadline Commerce faces for congressional written explanation.
  • 0, other commercial AI model APIs that have previously been subject to EAR export controls.

Why it matters

The suspension creates the first binding legal framework for treating AI model APIs as controlled exports. If the order survives Congress and courts, it bifurcates the global AI market by nationality: every future Anthropic frontier release will face the same regulatory question before deployment. The G7 "trusted partners" AI scheme, designed to restore allied governments' access to restricted US models, may become the only pathway through which non-US users can reach the highest-capability tier. The precedent also inverts the competitive dynamics: open-weight models from Deepseek and Meta face no equivalent nationality restriction, giving them a structural advantage in the non-US market.

What to watch

  • Commerce Department's written response to Congress by June 26.
  • Whether Anthropic challenges the directive legally (EAR challenges are rare; injunctions are rarer).
  • G7 "trusted partners" access negotiations, whether allied foreign nationals regain access first.
  • Whether Openai or Google Deepmind pre-emptively restrict their own frontier models to avoid a similar directive.