How is AI used in war today? Drones, targeting and who decides

How is AI being used in the military?

Most military AI is not in weapons. It sits in the back office and the intelligence cell, where the problem is too much data for too few people.

Intelligence: Project Maven (US, since 2017). In April 2017 the then US Deputy Defense Secretary, Bob Work, set up an "Algorithmic Warfare Cross-Functional Team" to work on what he called Project Maven. Its first job was computer vision: finding objects of interest in "millions of hours of video". The colonel running it said the goal was that "one analyst will be able to do twice as much work, potentially three times as much", and added that "AI will not be selecting a target [in combat] … any time soon".5

From finding objects to suggesting targets (US, 2024). Reported (Bloomberg). In February 2024, the chief technology officer of US Central Command, Schuyler Moore, told Bloomberg that algorithms developed under Project Maven had helped narrow down targets for more than 85 US air strikes in Iraq and Syria on 2 February 2024, and that the military began using them in actual operations after the 7 October 2023 Hamas attack. She said that in exercises, an AI recommendation engine "frequently fell short" of humans in proposing the order of attack or the best weapon to use, and that "every step that involves AI has a human checking in at the end".4

Everything else. The Congressional Research Service lists logistics, cyber operations, information operations, command and control and autonomous vehicles as other areas where AI was being built in, and notes that the Pentagon's AI centre began its first two "National Mission Initiatives" in 2019: predictive maintenance (fixing equipment before it breaks) and humanitarian assistance and disaster relief.14

Where it is heading (US policy). The June 2026 national security memorandum (NSPM-11) says the administration "will accelerate the development and use of AI for national security applications", and orders an update to the Pentagon's rules on autonomous weapons (more on those below).8

Do drones use AI?

Some military drones do, but usually for narrow tasks, and Ukraine is where this has gone furthest.

The most detailed public account we found is a CSIS study by Kateryna Bondar, published 6 March 2025 and based on interviews in Ukraine. What it found:1

• -What AI does: helps drones navigate, recognise and track objects, and finish the last part of an approach. CSIS says these tools are "still in their early stages of deployment", "limited, for now, to basic object recognition and tracking".
• -Who decides: "engagement decisions remain squarely in the human domain", and operators can "override autonomous functions".
• -Beyond drones: Ukraine's Delta system pulls video, photo, sound and text from many sources into one battlefield picture; CSIS treats it as the data backbone that future autonomy would need.
• -The goal: Ukrainian forces' long-term aim is "to maximize autonomy across the battlefield", with human input "limited to engagement decisions". CSIS also reports their objective of removing soldiers from direct combat and replacing them with autonomous unmanned systems.

So "AI drone warfare" today mostly means a person picks the target and software helps the drone get there. The question of drones that pick targets themselves, and the newer guidance kits and government tests in Ukraine, are covered on Are killer robots real?.

We found no source we could check on how Russia uses AI in its own drones, so this page says nothing about it.

AI-assisted targeting: what has been reported

The most debated use of military AI is software that proposes targets. Two cases have been reported in detail, and both rest on what officials or anonymous insiders told journalists.

Lavender (Israel, Gaza war). Reported (+972 Magazine and Local Call; shared in advance with the Guardian). On 3 April 2024, the journalist Yuval Abraham published testimony from six Israeli intelligence officers about an AI-based system called Lavender.3 As the Guardian summarised it, four of the sources said that early in the war Lavender listed "as many as 37,000 Palestinian men" linked by the system to Hamas or Palestinian Islamic Jihad. One source told the reporters: "I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added-value as a human, apart from being a stamp of approval."2

The Israeli military rejected that description. In a statement quoted by the Guardian, it said Lavender is a database used "to cross-reference intelligence sources" and "not a list of confirmed military operatives eligible to attack". It added that "the IDF does not use an artificial intelligence system that identifies terrorist operatives or tries to predict whether a person is a terrorist", and: "Information systems are merely tools for analysts in the target identification process."2 The +972 report also carries an IDF denial.3 We cannot check either account independently.

Maven in the Middle East (US, 2024). See above: a US official's own description, as reported by Bloomberg, of algorithms helping to find targets, with a human checking each step.4

What the two cases have in common: in both, officials say a person approves each strike. Where they differ is the claim, made by the Lavender sources, that the approval had become a quick formality. That is the centre of the next question.

Who decides? Human in the loop, on the loop, out of the loop

Three phrases come up in every debate about military AI:

• -Human in the loop: a person must choose and approve each target.
• -Human on the loop: the system can act on its own, but a person supervises and can stop it.
• -Human out of the loop: no one supervises the individual strike.

(Fuller definitions, with the US Defense Department's 2012 wording, are on Are killer robots real?.)

The official positions:

• -The US rule does not say that a human must approve every strike. The rule, DoD Directive 3000.09 (January 2023), says autonomous and semi-autonomous weapons "will be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force", and requires testing and a senior review before some systems are developed or fielded.6 It does not cover cyber tools or unarmed systems.6
• -The 2026 UN expert report, adopted by the states in the expert group (which works by consensus), says "control and human judgement" over lethal autonomous weapons "are needed", and that "accountability cannot be transferred to machines".10
• -The UN Secretary-General and the Red Cross go further: "The autonomous targeting of humans by machines is a moral line that we must not cross."12

The hard part is not the slogan but the practice. A person can be formally "in the loop" and still have too little time or information to disagree with the machine; the UN expert report names this risk, "automation bias", among the things states should guard against.10 The Lavender testimony ("a stamp of approval") is the best-known claim that this has already happened; the Israeli military denies it.2 US Central Command's account ("every step that involves AI has a human checking in at the end") is the opposite claim.4

For the highest-stakes version of the question, whether AI has any role in nuclear launch decisions, see Does AI control nuclear weapons?.

What are the rules?

• -International humanitarian law applies to every weapon, AI or not. The 2026 UN expert report restates that it "applies fully" to lethal autonomous weapons.10
• -US Directive 3000.09 (2023), described above.6 In June 2026, NSPM-11 ordered the Pentagon to issue an update "within 90 days", to be "reviewed annually".8 As of 16 October 2026, the Pentagon's own list of directives still shows the January 2023 version.7
• -The Political Declaration on Responsible Military Use of AI and Autonomy (US State Department, current text dated 9 November 2023) is non-binding. Endorsing states commit to measures such as legal reviews, use "within a responsible human chain of command and control", and the ability to disengage or deactivate systems "when such systems demonstrate unintended behavior". The State Department's own list names 58 endorsing states, including the US, "as of November 27, 2024"; it is the latest list the page shows.9
• -UN talks on autonomous weapons. In December 2025 the UN General Assembly backed further work on autonomous weapons by 164 votes to 6 (the US and Russia among the six), with 7 abstentions.11 The Geneva expert group adopted its final report on 4 September 2026 and asked states to "consider and endorse" its elements at a Review Conference on 16–20 November 2026, "without prejudging" whether they become a treaty.10 Reuters, citing a senior diplomat, reported that the US position had hardened, with Washington "favouring something more akin to guidelines than a legally binding international treaty"; Reuters said this position is supported by states including Russia, India and Turkey.13 The full ban debate is on Are killer robots real?.

AI in the military: pros and cons

Both columns draw on the same documents: mainly the US Congress's research service (CRS), the UN expert report and the UN–Red Cross call, plus the Maven and Reuters reports. Neither side's arguments are ours.

What are the dangers of AI in the military?

The risks named most often in the official and expert texts we opened, roughly from most immediate to most speculative:

1. 1.Mistaken targets and rubber-stamping. Software that suggests targets can be wrong, and a hurried human may not catch it ("automation bias").10 This is the core of the Lavender reporting, which the Israeli military disputes.2
2. 2.Unintended engagements and loss of control. Listed in the UN expert report adopted by the states in the expert group, including those that oppose a treaty.10
3. 3.Escalation and speed. CRS cites analysts who warn that "placing AI systems capable of inherently unpredictable actions in close proximity to an adversary's systems may result in inadvertent escalation or miscalculation".14
4. 4.An arms race. The fear that no country will slow down for fear of falling behind. CRS notes that Russia's president said in 2017 that "whoever becomes the leader in this field will rule the world"; NSPM-11 frames US policy as securing "a decisive and enduring AI advantage".148
5. 5.Spread to non-state groups. The UN report names "proliferation" and "acquisition by terrorist groups".10

The other side. Supporters argue that these are reasons to build military AI carefully, not to stay out: that adversaries will build it anyway (NSPM-11 says competitors deploy AI "with little regard for appropriate human oversight"),8 that it can keep soldiers out of danger,14 and that the law of war already governs it.13

How this connects to AI doom. These are risks of people using AI as a tool in war, which is different from the scenario on our Skynet page, where an AI pursues goals of its own. The two meet on one question: whether humans stay in control of systems that act faster than they can follow. For AI acting on its own outside the military, see Are AI agents dangerous?.

Should AI be used in the military?

This is a values question, and the sources split on it rather than on the facts:

• -"Yes, with rules": the US position. Use AI widely, keep "appropriate levels of human judgment", test and review it, and rely on existing law rather than a new treaty.6813
• -"Yes for some uses, never for others": the UN Secretary-General and the Red Cross call for prohibitions on some autonomous weapons and restrictions on all others, and say machines that take lives "without human involvement should be prohibited by international law".12 In December 2025, 164 governments voted for continued UN work on autonomous weapons.11
• -"Wherever it helps": Ukraine's stated long-term aim, under invasion, is "to maximize autonomy across the battlefield", while keeping humans for engagement decisions.1

Almost nobody in these sources argues for letting AI decide on lethal force with no human role at all. The disagreement is over how much human involvement is enough, and whether it should be a legal duty or a policy choice. The next marker is the UN Review Conference on 16–20 November 2026; this page will be updated after it.

FAQ

Why is AI in the military bad?

It isn't simply bad or good. The main worries are wrong targets approved too fast, systems behaving unpredictably, faster escalation and an arms race; the main arguments for it are fewer soldiers in danger and faster analysis. See pros and cons.

Is AI warfare already here?

Yes, in the sense that AI already sorts intelligence, suggests targets and steers drones in current wars.124 Not in the sense of machines fighting wars on their own.

Is ChatGPT or Claude used by the military?

Not covered here; we have not yet checked company and government sources on this.