The Alliance of the Algorithm

Europe is discovering that artificial intelligence is not simply another technology to regulate, but a new system of power in which dependence may matter more than principle

By David A. Williams

Artificial intelligence was supposed to be the technology that dissolved borders. Models would absorb knowledge from everywhere, developers would collaborate across continents and the resulting systems would become a common intellectual infrastructure, rather like the internet with opinions. That optimistic picture is now being replaced by something much harder. The United States increasingly treats its most advanced models as strategic assets whose capabilities must be protected from foreign extraction. China regards American restrictions as an attempt to preserve technological dominance under the language of intellectual property and national security. Europe, meanwhile, is trying to maintain commercial and scientific relations with both powers while insisting that AI should remain subject to laws written in Brussels. This is an increasingly uncomfortable position because the central dispute is no longer merely about how artificial intelligence should be governed. It is about who possesses the authority, computing power and military reach to define what governance means.

The latest confrontation concerns model distillation, a technique through which a smaller system can be trained using the outputs of a more capable one. American security agencies have accused Chinese companies including DeepSeek, Alibaba, Moonshot AI and Z.ai of conducting industrial-scale operations to extract restricted capabilities from leading American models. The allegations, reported by the Associated Press, go beyond the familiar complaint that Chinese firms have benefited from Western research. Washington says the companies used multiple routes to gain unauthorized access, breached terms of service and purchased premium subscriptions in bulk to reduce the cost of their operations. Beijing rejects the claims as groundless, arguing that distillation is a common industry practice and that the United States is attempting to establish a monopoly over AI. Neither account should be accepted uncritically. American companies themselves developed powerful models by ingesting vast quantities of human work, often without asking each writer, artist, publisher or programmer for individual permission. Washington is therefore defending a boundary between learning and theft that becomes considerably clearer when the alleged learner is Chinese than when the source material belongs to somebody else.

That contradiction does not make the American allegations irrelevant. It makes them politically revealing. The question is no longer whether machines can learn from other machines, since that is already happening, but which forms of technological imitation powerful states are prepared to tolerate. What was presented as an engineering method is becoming an instrument of foreign policy. Terms of service are beginning to acquire the strategic significance once attached to export licenses, while access to a commercial chatbot can become a question for intelligence agencies. Europe should pay close attention because it relies heavily on American platforms while also trading extensively with China. If Washington classifies more forms of model interaction as hostile extraction, European companies and research institutions may eventually be expected to demonstrate not only how they use American systems but with whom their knowledge, outputs and applications are subsequently shared. The digital supply chain will then become a chain of political custody.

This creates a peculiar problem for the European Union. Europe has attempted to distinguish itself through regulation, arguing that technological power can be made accountable through legal categories, transparency duties and protections for citizens. That approach has value, particularly when the alternatives are corporate self-policing or state secrecy. Yet rules do not automatically create capability. Europe may regulate access to systems whose decisive components remain designed, trained and operated elsewhere. It may specify how military and civilian AI should behave without possessing enough independent computing infrastructure, frontier models or operational experience to determine how those systems actually evolve. Strategic autonomy cannot consist solely of writing conditions for renting another power's machinery. A continent that depends upon American intelligence, American cloud capacity and American models while depending upon Chinese manufacturing and markets has not escaped the contest between Washington and Beijing. It has built its economy inside it.

The military dimension makes this dependency more serious. The Pentagon's chief digital and artificial intelligence officer, Cameron Stanley, has said that even America's closest allies lack the resources, experience and scale required to keep pace with the US military's adoption of AI. According to The Guardian, the Pentagon has rolled out an internal chatbot platform reportedly used by 1.7 million personnel and developed systems intended to shorten the interval between gathering intelligence and making a command decision. Stanley presented speed as the central advantage, describing the ability of commanders to act without passing through the traditional sequence of briefings and formal reviews. From an operational perspective, that may look revolutionary. From a democratic perspective, it raises a less comfortable question: when technology accelerates a military decision, which parts of the decision-making process has it made more efficient, and which safeguards has it simply removed?

For smaller NATO members such as Denmark, the widening capability gap presents a strategic dilemma that cannot be solved merely by increasing defense expenditure. Denmark can purchase advanced platforms, join shared programs and contribute excellent specialists, but it cannot reproduce the American military AI ecosystem on a national scale. It will therefore rely upon systems whose training data, architecture and operational assumptions may be largely determined in the United States. That reliance may be unavoidable, but it should not be mistaken for technological partnership between equals. If an American system ranks threats, combines intelligence or recommends courses of action, Danish forces may gain speed while losing the ability to examine fully how the judgment was constructed. The danger is not that the machine will suddenly seize command. It is that human commanders, operating under extreme pressure, will find it increasingly difficult to reject a recommendation supported by an apparently superior system belonging to the alliance's dominant military power.

This is where the confrontation over Chinese model distillation and the Pentagon's warning about allied weakness become parts of the same story. In the civilian debate, the United States is attempting to stop a rival from extracting capabilities from American systems. Within the military alliance, it is offering partners access to capabilities they cannot independently reproduce. The first relationship is described as theft or exploitation; the second as cooperation and interoperability. Yet both demonstrate the same underlying reality: the state that controls the most capable models can decide who receives access, on what terms and for which purposes. AI governance is therefore becoming inseparable from hierarchy. Europe may be America's ally, but alliance does not abolish dependence. It merely makes dependence politically acceptable, at least until the interests of the supplier and the user begin to diverge.

Denmark should respond neither with technological nationalism nor passive trust. Building a Danish equivalent of the Pentagon's AI infrastructure would be unrealistic, wasteful and probably impossible. The more credible objective is to preserve sovereign judgment within shared systems. Danish authorities should demand meaningful access to testing, audit procedures and operational limitations before adopting AI that influences military decisions. Procurement agreements should protect the ability to operate, inspect and, when necessary, refuse automated recommendations. Europe should also invest collectively in secure computing capacity, defence-specific evaluation and models that can function under European legal authority. The essential capability is not to duplicate every American platform. It is to ensure that Europe retains enough knowledge and infrastructure to recognize when its interests no longer coincide perfectly with those of its provider.

The geopolitical contest over artificial intelligence will not be settled by discovering a universally agreeable definition of responsible innovation. China, the United States and Europe approach the subject from different positions because they possess different forms of power. Washington has frontier companies, immense computing capacity and unparalleled military reach. Beijing has scale, industrial depth and a political system capable of directing national resources towards strategic technologies. Europe has a vast market, regulatory influence and a habit of believing that rules can tame forces it does not fully control. That belief will now be tested. The uncomfortable lesson for Denmark and its European partners is that sovereignty in the age of AI will not mean owning every model or severing every dependency. It will mean knowing which dependencies can be managed, which must be reduced and which quietly transfer the power to decide into somebody else's hands.

Sources

Associated Press: “China hits back at US claims of 'malicious' AI distillation ahead of planned Trump-Xi talks”, September 9, 2026.

The Guardian: “US allies lack resources to keep pace on AI, top Pentagon official says”, September 9, 2026.