By Stuart Kerr, Technology Correspondent, LiveAIWire
AI military strategy crossed a visible threshold during the war in Iran in March 2026, when the Pentagon’s Maven Smart System helped US forces strike more than 1,000 targets in the first 24 hours of the conflict. That was a tenfold increase over what was achievable before the platform existed, according to a detailed CSIS analysis of the Department of Defense’s own targeting data. AI is no longer a supporting tool sitting behind military decision-making. It has become the interface through which targeting decisions are made, the layer that decides what a commander sees first, and increasingly the system that recommends which weapon should strike which target and when.
The ethical questions raised by AI military strategy are not hypothetical, and they are not being avoided because nobody can articulate them. The United Nations has negotiated on a framework for autonomous weapons since 2014 under the Convention on Certain Conventional Weapons. In 2025, Secretary-General Antonio Guterres set 2026 as the deadline for a legally binding treaty, calling systems that select human targets without human control politically unacceptable and morally repugnant. More than a decade of talks over AI military strategy and autonomous weapons have produced no binding agreement.
The reason is not that the questions are unanswerable. It is that the countries building the most capable systems are not yet willing to accept the constraints that answering those questions honestly would require.
What AI Military Strategy Actually Looks Like Inside the Pentagon
The clearest evidence of how far AI military strategy has moved inside the Pentagon is the Maven Smart System, the flagship software platform Palantir built on top of the Pentagon’s original Project Maven initiative. In March 2026, Deputy Secretary of Defense Steve Feinberg designated the system a formal program of record. That gave it stable long-term funding and moved its oversight to the Chief Digital and AI Office as the intended cornerstone of the Pentagon’s joint command and control strategy. The system now fuses satellite imagery, drone video, signals intelligence and battlefield sensor feeds into a single interface that lets a small team do what used to require thousands of staff.
A detailed CSIS analysis of the Maven Smart System found that a 20-person targeting cell using the platform matched the output of the 2,000-person cell that ran time-critical targeting during Operation Iraqi Freedom, widely regarded as the most efficient targeting operation in US military history. That comparison, roughly a hundredfold reduction in the staff needed to run a targeting cell, shows how far AI military strategy has already reshaped what a modern military organisation looks like, long before any autonomous weapons treaty has been agreed.
Large language models add a further layer on top of that computer-vision foundation. Anthropic’s Claude became the first frontier model cleared for use on classified US military networks, through a 2024 partnership with Palantir. Pentagon officials cited by CSIS say natural-language analysis on top of existing detection systems produced a further fivefold increase in targeting speed. That relationship has since deteriorated. Anthropic refused Pentagon demands in 2026 to strip the terms of service restricting its models from fully autonomous lethal weapons and mass domestic surveillance, and the Pentagon responded by labelling the company a supply chain risk, a dispute Anthropic is currently contesting.
The episode is a live example of the tension at the centre of AI military strategy. The military value of a targeting system rises with its autonomy, and the safeguards a responsible AI developer wants to keep in place fall as that autonomy increases. No framework currently in force resolves which side of that trade-off should win.
The Autonomous Weapons Question the UN Has Not Resolved
The International Committee of the Red Cross defines an autonomous weapon system as one that selects and applies force to targets after activation, without a human choosing the specific target or the moment of attack. The ICRC has urged states since 2015 to adopt legally binding rules restricting these systems as part of any credible AI military strategy framework. It argues that the loss of human judgement over life-and-death decisions raises humanitarian, legal and ethical concerns that voluntary guidance alone cannot resolve.
Current tracking by the Stop Killer Robots coalition found that as of early 2026, roughly 127 countries support some form of binding ban or restriction. Twelve countries, including the United States, the United Kingdom, Russia and Israel, explicitly oppose a ban, with around 53 nations yet to state a clear position. That split is the practical shape of the AI military strategy debate at the UN: broad numerical support for restriction, concentrated among states with the least capacity to build these systems, against a small bloc of the most militarily advanced states that oppose being bound.
The definitional argument at the centre of the UN process is what counts as meaningful human control. Washington’s position has generally been that human oversight of a mission as a whole satisfies the requirement, even when no human reviews each individual targeting decision the system generates. The ICRC and most of the civil society coalition working on the issue argue that mission-level supervision is not sufficient for lethal decisions. International humanitarian law requires that distinction between combatants and civilians, and the proportionality of expected harm, be judged in the context of each individual strike, not approved in advance as a blanket policy.
Why Speed Is the Real Driver of AI Military Strategy
The military case for autonomy is fundamentally a case about speed. Missile defence systems, including the Navy’s Phalanx close-in weapon and Israel’s Iron Dome, have used automated engagement authority for years, because human reaction times are too slow for some interception windows. The logical extension, from automated interception of incoming munitions to autonomous engagement decisions involving other kinds of targets, is shorter than the definitional debates at the UN suggest. That is why military planners in the US, China and elsewhere treat decision speed as a core strategic asset of AI military strategy rather than a side effect of automation.
That same speed dynamic is reshaping cyber operations, where the absence of clear international law creates even more uncertainty than the physical battlefield. LiveAIWire’s coverage of the Five Eyes joint warning on AI-enabled cyberattacks found that the cybersecurity chiefs of the United States, United Kingdom, Australia, Canada and New Zealand concluded frontier AI would reshape the offensive cyber landscape within months rather than years. That timeline applies as directly to military cyber operations as it does to criminal ones. When an AI system can identify a vulnerability, write exploit code and act on it faster than a human chain of command can be briefed, the assumption that a human always retains a meaningful decision window starts to break down.
The Governance Gap Behind AI Military Strategy
The UN process has produced statements of concern but no binding treaty, for a reason that is straightforward once the incentives are laid out plainly. The countries with the most advanced military AI programmes, principally the United States, China, Russia and Israel, are also the countries whose support any binding treaty would require to be meaningful, and none of them have signed on to a ban. China has called for a treaty restricting the use of lethal autonomous weapons while continuing to develop them, a position Human Rights Watch has described as consistent with China’s own advanced weapons programme rather than a genuine concession.
The result is a governance gap in AI military strategy that has persisted for more than a decade. It is not that the ethical and legal analysis is unfinished. It is that none of the states capable of building these systems are willing to accept the trade-off that binding restriction would impose while their rivals continue building.
That same asymmetry of control now extends beyond weapons systems into the AI models themselves. LiveAIWire’s coverage of the AI export controls that shut down Anthropic’s most advanced models worldwide in June 2026 showed that a single Commerce Department directive can remove a piece of critical AI infrastructure from allied nations overnight. Governments now treat that as a standing feature of relying on any single country’s AI supply chain, not a one-off dispute.
The same logic that lets Washington restrict export of a commercial AI model for national security reasons underpins the wider contest over who controls the most capable military AI systems, a dynamic LiveAIWire’s reporting on the AI Cold War between the United States and China has tracked through a year of shifting chip export policy.
What This Means for Anyone Watching AI Military Strategy Unfold
For organisations and citizens outside any defence establishment, the practical significance of AI military strategy is that the norms being set now in classified programmes and UN conference rooms will define how AI-enabled force is used for decades. The models powering Maven Smart System’s targeting recommendations are close relatives of the commercial AI systems used across finance, healthcare and journalism. That means the access controls, safety commitments and export restrictions being negotiated in a military context increasingly shape what is available in every other context too.
LiveAIWire’s coverage of how state actors are using AI for espionage and cyberattacks found the same vetted access perimeter, built around a handful of US firms and allied governments, that now governs the most capable military AI systems. The pattern repeats across every domain where frontier AI capability intersects with national security: a narrow circle of states and companies holds the most powerful tools, while the rules for using them responsibly remain unwritten.
The International Law Question That Has Not Been Settled
International humanitarian law requires that every attack comply with the principles of distinction between combatants and civilians, proportionality between expected harm and military advantage, and precaution to minimise civilian harm. These are judgements that international law has traditionally required humans to make and be held accountable for. The ICRC’s position, supported by a growing body of legal scholarship, is that fully autonomous weapons systems cannot reliably make these context-sensitive judgements. It argues they should be prohibited outright when designed to target human beings directly.
Governments developing these systems argue instead that specific systems can in principle be validated against these standards. That shifts the debate in AI military strategy from a categorical prohibition to a case-by-case certification problem that no state has yet agreed to submit to independent verification. Whichever position prevails will determine whether the current generation of AI-enabled targeting systems can be lawfully deployed under existing international law, or whether an entirely new legal framework is required.
Guterres has set 2026 as the year that gap should close. Whether it does depends less on the strength of the legal argument, which is already well developed, than on whether the states with the most advanced AI military strategy programmes are willing to accept limits on capabilities they currently consider a decisive strategic advantage.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.