When the US Commerce department issued a directive requiring AI leader Anthropic to cut access to its Fable and Mythos models for all foreign nationals, Europeans feared the worst about Washington’s power to fire a “kill switch.”
In reality, their fears were overblown. The shutoff represented a kneejerk response to an alleged jailbreak, using the only tool available to the administration: export controls. If the White House wanted to coerce its allies, it has other, less self-destructive ways to do so than ordering its most profitable companies to shut off their services.
Despite this, a counterproductive kill switch meme drives European policy debate. US policy that imposed export controls on AI chips under both the Biden and Trump administrations has fueled support to de-risk from the US. Europe’s kill switch fear jeopardizes Washington’s vision of a US-led global AI stack to counter China. The US could and should reassure Europeans by guaranteeing access to US AI technology.
Washington recognizes the need for European support to break China’s stranglehold on critical minerals. Its Pax Silica initiative aims to build a western AI supply chain alliance. The European Union officially joined Pax Silica in late June after France lifted its objections, having received informal guarantees that membership would not require the EU to mirror US chip export controls, for example, on ASML’s semiconductor manufacturing equipment.
But the EU received no guarantees, even informal, of future access to advanced US chips or AI models. When the EU’s most senior digital policy official, Roberto Viola, visited Washington to sign the Pax Silica declaration on behalf of the EU, such guarantees were his number one ask. He returned to Brussels empty-handed. The US and Europe also discussed a “trusted partner” scheme at this year’s G7, but US controls on frontier models have only tightened since then.
US officials are unrepentant. In a recent blog post, US State Department Under Secretary Jacob Helberg dismissed Europe’s tech sovereignty aspirations as backward and counterproductive. He did not address the reasons why Europe is pursuing tech sovereignty.
Washington wants to retain control: loyalty from allies to a US-dominated tech stack, while reserving the right to leverage its dominance when convenient. In this context, it is understandable that the EU and Canada are looking to band together to strengthen their bargaining power and explore alternatives to the US AI stack.
Administration officials understand the risks of inadvertently pushing buyers to find alternative suppliers. In 2025, the administration relaxed export controls on NVIDIA chips to China, on the basis that squeezing Beijing’s access would only cede market share to its homegrown alternative, Huawei.
The timing is unfortunate: these kill switch fears come as Europe has belatedly caught up with the US in acknowledging the security risks of Chinese tech products. Over the last few years, the EU has shifted its approach to China, moving from treating it as a traditional trading partner to classifying it as a “systemic rival” and passing or proposing several key pieces of far-reaching legislation aimed at protecting European markets. French and German leaders are now pushing to impose a tool similar to the US’s Section 301 tariffs that would allow them to close entire markets to Chinese goods.
Despite this recognition, the allure of Chinese open-weight models such as Qwen, DeepSeek, and Kimi may prove irresistible. These models are comparable in performance to their US counterparts, yet are much cheaper to run. Crucially from a sovereignty perspective, they provide alternatives to leading US AI developers, reducing the latter’s global market power, and they can be freely downloaded and run locally — not shut off remotely from the cloud.
If Washington is serious about maintaining tech leadership, it should offer a guarantee to Pax Silica allies: the US will not deny access to the AI stack, whether chips, models, or cloud. Where genuine security concerns exist around frontier model capabilities, this access guarantee would take the form of a commitment to equal treatment of allies and joint governance.
Skeptics will point out that the current Administration’s decisions can change, and that the Biden Administration’s AI diffusion framework first introduced the idea of withholding advanced AI chips from European allies.
Both objections are valid. But a gesture of reassurance would go a long way to rebuild trust and defuse the current sovereignty debate — and allow Europe to move on from counterproductive talk of kill switches.
Nick Reiners is a senior analyst in Eurasia Group’s Geo-technology practice, specializing in European tech policy. Previously he was a civil servant and diplomat at the UK Representation to the EU in Brussels.
Bandwidth is CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy. All opinions expressed on Bandwidth are those of the author alone and may not represent those of the institutions they represent or the Center for European Policy Analysis. CEPA maintains a strict intellectual independence policy across all its projects and publications.
Steel and Silicon: Allied Innovation for the Next Century
2026 CEPA Tech & Security Conference
