A Race Without Rules
While states negotiated principles for military AI in Spain, on the battlefield the operator's signature had already shrunk to a formality of seconds.
On 4 and 5 February 2026, eighty-five delegations gathered in A Coruña for the third summit on responsible artificial intelligence in the military domain, and when it closed the outcome document, Pathways to Action, carried barely thirty signatures (the summit’s official site collects the proceedings). Washington and Beijing stayed off the list, a retreat from Seoul, where two years earlier some sixty states had endorsed the Blueprint for Action. Diplomacy had set out to turn principles into binding constraints; the result measured how little they weigh against the calculus of power. The Hague in 2023 opened the process with a broad, undemanding statement of intent, Seoul tightened the language, A Coruña exposed the limit, because every step towards binding force narrowed the circle willing to commit.
The retreat follows from the stakes. States understand the systemic risks of handing decisions to machines, yet hesitate to bind themselves first, since a country that accepts a constraint its rivals ignore opens a gap it cannot close. International relations theory has named this friction for decades: the shared interest in a control regime coexists with each player’s incentive to defect, fed by the suspicion that others already have. The three states that matter confirm the rule. The United States promotes multilateral guidelines while guarding its freedom of action, having built around its 2023 Political Declaration a perimeter that binds others more than itself. China runs a parallel line, indulging the dialogue tables while shielding its modernisation from any external limit. Russia pushes the logic furthest, because a high-intensity war and its dependence on autonomous and semi-autonomous systems turn any regulation into immediate strategic loss.
The contest then moves to the technical ground of definition, where terminology becomes competition for advantage. Deciding where algorithmic assistance ends and autonomy begins decides which technologies fall under limitation and which escape it, and every line drawn penalises one state’s modernisation while sparing another’s. Rigid categories break against the fluidity of the technology: automation runs along a spectrum, from machines that act in full detachment from human input, to platforms that take strategic direction and run their own tactics, to architectures that leave command only the gesture of ratification. The same question has kept the UN’s own track, the Group of Governmental Experts on lethal autonomous weapons in Geneva, circling for years without convergence. While diplomacy stalls on taxonomy, the technologies are validated elsewhere, at the front.
The front as laboratory
The wars under way are testing grounds for these doctrines, and they mature them faster than any forum. In Ukraine the fusion of algorithmic systems, surveillance networks and drone swarms has pushed decision-making down the chain of command, towards the machine. In Gaza, through the testimony of Israeli intelligence officers gathered by +972 Magazine, the war exposed the platforms built for the mass processing of information: Lavender generated a list of some thirty-seven thousand names, Habsora selected buildings as targets, Where’s Daddy? tracked the marked to the moment they reached home. Officers reported treating the machine’s output as a human decision, the check often running to the twenty seconds needed to confirm the target was male before signing the strike order. Speed of computation now decides an operation’s effectiveness, and AI has reshaped how commanders behave, drawing them to validate the algorithm by inertia until human control slides from review to ratification.
The collapse has a name in the study of automated systems. The literature on automation bias has shown for years how a system’s prolonged reliability erodes the operator’s scrutiny: the machine’s correctness becomes a given, and vigilance over its recommendations falls, an effect found in novices and experts alike that training does not cancel (the standard synthesis remains the one published in Human Factors in 2010). In a cockpit or a control room that slackening causes accidents; in the military domain it causes strategic asymmetries, as a software error becomes a lethal collateral, a diplomatic crisis, an escalation command struggles to govern. The contradiction holds, between the speed of the decision and the weight of responsibility: technology compresses the cycle of action, while strategic deliberation demands context, qualitative judgement, validation that resists quantification.
Automating the decision cycle
The known flaws of these models in civilian use do not vanish on the way to defence; they pass whole into military systems, where dataset bias, classification errors, the perishability of information and exposure to cyber attack change scale. In a market the algorithmic error costs money; in war the same deviation destroys the wrong target and opens a spiral between sovereign powers. The threat is automation taking root across the entire decision cycle, as the systems redefine intelligence gathering, threat mapping, target assignment, logistics and damage assessment in turn, shifting military effectiveness onto the handling of algorithmic flows and making command structurally dependent on the machine’s suggestions.
The political contest of the coming decades arrives without spectacle and thick with doctrinal complexity, and its knot is how much decision-making commands choose to keep in an environment that rewards speed and informational superiority. Competition forces rising degrees of delegation rather than the loss of the edge, and the history of war records, without exception, the tendency of states to absorb any innovation that raises lethality or efficiency. Artificial intelligence follows the same path. Whether international institutions can build stabilising mechanisms that blunt the race, and keep diffusion from outrunning the diplomatic capacity to regulate and manage crises, will measure the distance between the principles of A Coruña and their application.