18 Days, One Order: How a US 'Kill Switch' Cut Europe Off From Frontier AI

For 18 days in June 2026, the United States quietly pulled the plug on Europe's access to Anthropic's most advanced artificial intelligence. The order, dated June 12, came from the U.S. Commerce Department and instructed Anthropic to block all foreign nationals from using its frontier models ⁹. When the restriction was lifted later in the same month, the technical access came back. The political alarm did not ⁶⁷.

The episode has been described by Al Jazeera as a wedge driven between the United States and its allies ³. UPI framed it more bluntly: whether the legal authority existed or not, the U.S. government had shown it could control global access to the world's most powerful AI ⁹. The Verge argued the shutdown handed momentum to the global sovereign-AI movement ². And on July 22, Secretary of State Marco Rubio directed American diplomats abroad to play down the very phrase — "kill switch" — that had come to define the controversy, telling foreign governments in a Reuters exclusive that American technology contained no such thing ⁶⁴.

The June 12 instruction was sweeping in scope and sudden in execution. Anthropic, one of the leading U.S. AI laboratories, was told to cut foreign nationals off from its newest systems. The Verge's reporting identified the affected models as those the publication referred to as "Fable and Mythos," with the caveat that full technical detail sat behind truncated text ². No European regulator was consulted. No European government was given advance warning. The decision traveled one way: from Washington outward, with no off-ramp for the companies, universities and public services on the receiving end ³.

Reporting is consistent that the restriction was short-lived and ended later in June, and HeyDay's own coverage described an 18-day window. The 18-day figure does not appear verbatim in any of the written sources cited here; sources ⁶ and ⁷ confirm brevity without naming a duration. Readers should treat the precise number as plausible but not independently verified by the documents in hand.

The country that reacted loudest was not Brussels but Berlin. Thomas Jarzombek, Germany's digital state secretary, argued in a DW interview that Europe has a dependence problem on foreign AI and needs more tech sovereignty. "Tech sovereignty is straightforward," he said ¹⁰¹¹. The DW-derived reporting confirms the interview and his core argument. The stronger phrase "wake-up call for European tech sovereignty" — circulated in coverage of his remarks — does not appear verbatim in the sourcing provided here and should be treated as a paraphrase rather than a direct quote.

Beyond rhetoric, Germany began pursuing tech sovereignty through procurement policy, favoring European AI providers, and by expanding domestic data center capacity to host and train advanced models at home. The specific measures described in our own broadcast sit in a thin evidentiary zone: the DW-derived sources confirm Jarzombek's tech-sovereignty frame but do not itemize the data-center build-out. Computerworld reported separately that Europe is looking to harden its legal position against any forced shutdown of AI, treating the June episode as the moment a hypothetical "kill switch" became a likely one ⁸.

The episode detonated at an awkward moment: the G7 AI summit in mid-June. Euronews reported that the ban confirmed the very "kill switch" fears Western governments had been carrying into the meeting, and supercharged European tech-sovereignty efforts ⁵. Foreign Policy, writing on July 20, argued the ban did not amount to a coherent AI policy and called instead for "an actionable agenda that ensures safe release and stable access" ⁷. Forbes ran the line the same day in its headline: "America Says Its AI Has No Kill Switch, But The World Is Hedging" ⁴.

That hedging is the political residue. Even after access was restored, even after Rubio's reassurance campaign, the precedent was set. A single executive-branch order had, in effect, redrawn the boundary of who could use frontier AI — and the boundary had been redrawn by Washington, not by the users on the other side.

Three things, each with a different shelf life. The technical fix — restoring foreign access — happened within weeks ⁶. The diplomatic repair is underway but fragile, with the U.S. now publicly contesting the framing its own order created ⁴⁶. The strategic shift is the slowest and the most consequential: Europe, and Germany in particular, are now treating frontier-AI access as a sovereignty question rather than a procurement question ⁸¹⁰¹¹.

That reclassification matters. Sovereignty questions get budgets, laws and industrial policy. Procurement questions get tenders. The June episode nudged the European debate from one column to the other, and the column it landed in is harder to walk back.