Brussels engaged in discussions with Washington concerning Anthropic's access following the Fable 5 prohibition.

      Brussels has initiated discussions with Washington regarding Anthropic after a US export-control order severed Europe’s access to the company’s two most advanced models, which included ENISA. To acquire an American AI model, the European Union finds itself negotiating not with the manufacturer but with the government that can disable it.

      Following a US export-control order in mid-June that abruptly restricted European access to Anthropic's most sophisticated systems, Brussels engaged in talks with the White House. The critical aspect to consider is the chain of command: the EU's pathway back to the technology passes through Washington, rather than through Anthropic.

      This cutoff occurred on June 12, when the US Commerce Department, in a letter from Secretary Howard Lutnick and prepared by the Bureau of Industry and Security, directed Anthropic to halt access to its Fable 5 and Mythos 5 models for any foreign nationals, both within and outside the United States, including the company’s own non-citizen employees. Anthropic had to disable both models globally because it couldn't verify nationality within the shared cloud infrastructure. However, its less powerful Claude models, including Claude Opus 4.8, remained unaffected.

      The stated reason for this action was a jailbreak. Officials responded after discovering a method to circumvent the models' safety measures, reportedly highlighted by researchers at Amazon, Anthropic’s largest investor. Anthropic contested the extent of the response, claiming that the jailbreak was limited to unlocking a specific Mythos cybersecurity feature in one particular instance rather than broadly bypassing Fable 5’s safeguards, thus not warranting such widespread access restrictions.

      The timing was particularly painful for Europe—just days before, Anthropic had been in discussions to grant the EU’s cybersecurity agency, ENISA, access to Mythos, marking the first opportunity for the model to be utilized outside the US and the UK. Various European governments, companies, and research institutions, including ENISA and NATO, that had previously gained access through Anthropic’s Project Glasswing were abruptly cut off without prior notice or a timeline for reinstatement.

      The order affected allies as well. Downing Street approached the White House requesting a UK exemption, only to be met with a definitive “zero chance” response, clearly indicating the limited leverage foreign capitals hold in negotiating access to a privately-owned American product.

      This scenario set the stage for the EU's own approach towards Washington. The political response across Europe was swift and introspective. French leaders urged Paris to expedite support for Mistral, the EU’s primary contender in frontier-model development. This incident reinforced a broader viewpoint that reliance on American technology poses a strategic risk rather than merely a procurement situation. A model that can be restricted by another government’s decree is not infrastructure under one’s control.

      Recently, signs of a thaw in the standoff have become apparent. Anthropic CEO Dario Amodei met with Donald Trump at the G7 summit in France, which seemed to soften the president's position, as Trump later remarked that he no longer considered the company a national security threat. Whether this leads to restored access for Europe and under what conditions remains an open question. For now, the EU has discovered that the control switch is not located in Brussels.