
In June the British government published its Strategic Defence Review (SDR), a document setting out the future of the armed forces over the next decade. Unlike many of its predecessors, which often served as bureaucratic compromise between Treasury caution and military ambition, this one trumpeted something new. Defence Secretary John Healey declared that artificial intelligence (AI) would be at the centre “of a new way of war”. To make good on that promise, the Treasury earmarked £4bn in the Spending Review, explicitly for autonomy. Britain’s defence establishment has long spoken of modernisation, but it has been made unequivocally clear that code, not just iron and steel, would lie at the heart of its future military strength.
The Ministry of Defence (MoD) has been edging in this direction for years. In 2021 the Royal Navy demonstrated an experiment with MADFOX, a crewless vessel deployed in British waters. The trial involved MADFOX working alongside a Puma drone, which gathered intelligence. The data was then used to direct a test missile launch. Though modest in scale, the trial was symbolically important. For the first time in a public exercise, Britain had shown that the loop of sensing, analysis and strike could be performed largely by machines, with human oversight kept deliberately at a distance. The Navy insists that humans remain in control, but the demonstration revealed how far autonomy could be taken if required.
The following year the Army staged its own showcase. Two platforms of unmanned aerial systems (UAS) were put forward: one developed by Atlas, the other by Elbit, an Israeli defence company with extensive experience of drones honed in the hellscape of Gaza. Elbit’s system stole the limelight. It demonstrated how one soldier could task six drones at once, each communicating with the others, planning their own missions and identifying points of interest for further scrutiny. This was not remote-control flying but autonomous coordination. A report from the House of Lords’ committee on AI in weapon systems, importantly titled Proceed with Caution, noted that neither system had been armed, but the implication was clear: Britain’s military was moving towards a world in which lethal acts could be delegated to algorithms.
For the most part, however, official pronouncements since then have stressed less controversial uses. Most initiatives are described as data analysis or logistics, fields where the prospect of lethal force is distant. Where autonomy has been introduced, it has usually been confined to what the Lords’ committee called “lower cluster” environments: bomb disposal, surveillance, or maritime patrols far from civilians. A robotic dog, developed for explosive ordnance disposal, is the most benign example.
The “SMASH” weapon sight, by contrast, is unashamedly lethal. Mounted on the Army’s SA80 rifle, it uses AI to detect and track a target, automatically permitting fire only when the crosshairs align with its digital kill box. The manufacturer claims this raises the probability of a successful hit to between 70% and 80%, compared with far lower figures for ordinary marksmanship. Here AI does not remove the soldier from the decision, but it claims to remove much of the soldier’s fallibility (though the AI sight cannot distinguish between friend and foe, just yet).
The Royal Air Force (RAF) has gone further still. Its Tempest programme, the planned replacement for the Typhoon fighter, integrates algorithms not merely to process data but to learn from its pilot. The system can adapt to a pilot’s behaviour, anticipate actions and, if required, fly independently.
Officials are quick to insist that pilots will not be replaced but enhanced, with AI acting as a second crew member rather than a substitute. Yet the tone of official commentary has shifted. AI is now spoken of as an integral part of aircraft design, not an experimental bolt-on. And with war raging in Ukraine, the rhetoric has hardened. The RAF has announced StormShroud, a drone designed to disrupt enemy air-defence networks using a high-tech BriteStrom jammer. Britain has invested £19m in the project. Though details have not been made public, it is widely believed that StormShroud will be able to conduct swarming operations, flying in coordinated groups without central human control. That would represent a leap beyond augmentation into the realm of AI-directed combat.
Such projects are possible only because of industry. The British armed forces are no longer producers but customers. In 2023–24 the MoD spent £37.6 billion on industry contracts, an increase of £4.6bn from the previous year. The sum covers everything from food to frigates, but it demonstrates how dependent the forces are on outside suppliers. AI is a small but growing portion of this spending.
Tussell, a contracts-tracking firm, calculates that the MoD is the UK’s third-largest government body in terms of AI contracts, with £272m awarded since 2018. The recipients range from defence stalwarts to Silicon Valley firms. BAE Systems, the country’s biggest defence company, features heavily. Adarga, a British start-up focusing on natural language processing, has also won repeat work. Frazer-Nash, QinetiQ and Thales appear too. But alongside them stand American names: Anduril Industries, IBM, Lockheed Martin and L3Harris.
Palantir Technologies looms largest. The Denver-based company has won every central government contract for data analytics. Its systems now underpin Britain’s defence data flows. Palantir has thus embedded itself deeply in the machinery of the state. Its proprietary frameworks have become the skeleton of government data handling. Switching to another provider would be disruptive and costly, giving Palantir extraordinary leverage. The imbalance is stark. Since 2018 American firms have received £107.5m in AI-related MoD contracts. British companies, by comparison, have been awarded just £21m. Britain’s dependence on American hardware has long been obvious in nuclear submarines and combat aircraft. The risk now is subtler: dependence on American software, which shapes how Britain’s military understands and acts on information.
Universities, too, have been drawn into this new military-industrial complex. Imperial College London told AOAV that it has received nearly £5.7m for four projects: insect-inspired autonomous navigation (one can only guess being designed for drone swarms); trusted software verification using deep learning; scalable AI simulation frameworks; and machine learning for software-defined networks.
The University of Edinburgh disclosed following our Freedom of Information request it had three contracts worth £97,442: one on sonar, another on machine-learning models, and a third withheld on security grounds.
Other universities, asked whether they had received MoD contracts, invoked national-security exemptions and declined to answer. The MoD itself is cagey. Freedom of Information requests are met with references to “operational sensitivity”.
Officials direct questioners to the Defence AI Playbook, a glossy document that lists areas of interest such as large language models, resupply logistics and object detection in satellite imagery, but avoids naming any actual systems.
Civil-society groups have tried to prise open the curtain. Drone Wars reported in 2023 that FOI requests revealed 73 R&D initiatives involving AI. Most concerned data analysis, data sharing or uncrewed vehicles. Few involved weaponisation. The RAF disclosed that it was testing predictive maintenance on its A400M transport aircraft, using sensor data to forecast repairs. Such applications are useful but are not at the apex of concern. They reduce workloads and increase efficiency but do not make headlines. Thankfully, the pattern that emerges is of AI being deployed cautiously, in ways that supplement rather than supplant humans.
Yet the potential for escalation is there, and it is growing. Of course, by global standards, Britain remains a modest player. The private sector is dominated by the US and China.
The Brookings Institute, a Washington think-tank, reported in March 2024 that the Pentagon’s AI contracts rose from 254 in 2023 to 657 in 2024. Their value ballooned from $269m to $4.32bn. The Pentagon accounted for 95% of all federal AI spending. Deals worth tens of millions are now routine. Anduril secured an $86m contract to provide autonomy software to Special Operations Command. Firestorm Labs won a $100 million contract for drones built with 3D-printed components, designed for fast parts replacement on the battlefield. Britain, by contrast, has spent just £937 million on AI contracts since 2018. America is also shifting from small experiments to large, programmatic contracts intended to embed AI across its forces.
China has chosen a different route. Its drones are comparatively affordable and increasingly effective. CH-4 and Wing Loong 2 UAVs sell for $1 – $2m. America’s Predator and Reaper cost between $4m and $16m, price tags three to five times higher. The gap is stark and persistent. For states unwilling or unable to pay American prices, Chinese exports are attractive.”
Russia, with less industrial depth, has pursued state-directed AI development, though with limited results. Europe lags. BAE Systems is the continent’s only company with scale to compare, but even it is dwarfed. EU efforts at joint AI projects remain slow, hobbled by national rivalries. France and Germany continue to quarrel over joint aircraft programmes. Post-Brexit Britain, outside EU frameworks, is more dependent than ever on American partners.
One unlikely pioneer is Ukraine. Since 2014, when Russia annexed Crimea, Ukrainian coders and engineers have improvised technologies to meet battlefield needs. The Delta System, integrating real-time data from multiple sources, is one example. Commercial drones have been modified to drop explosives. Civilian software has been adapted for military use.
Analysts such as Michal Onderco draw three lessons from Ukraine’s experience. First, the line between civilian and military technology is disappearing. Second, almost anything can be weaponised. Third, autonomy is not a luxury but a necessity for those with limited resources. AOAV would add a fourth: in the rush to defence, the protection of civilians is often secondary to the AI debate.
For Britain, which has focused on high-end platforms like Tempest, Ukraine’s improvisation is a reminder that agility and cost-effectiveness may matter as much as technological sophistication.
The expansion of autonomy has provoked a scramble to draft ethical frameworks. America’s Department of Defense published principles in 2020. NATO, Britain and Australia have followed with their own, emphasising accountability, bias mitigation, explainability, reliability and lawfulness. Yet such principles are easy to sign and hard to implement.
The war in Gaza illustrates the gap between aspiration and practice. In 2024 +972 Magazine, an Israeli outlet, reported that the Israel Defence Forces were using AI systems named Lavender and Gospel to generate huge target lists with minimal human oversight. Lavender was described as an “AI-driven kill list” with more than 30,000 names. Gospel was credited with generating up to 100 targets a day, compared with 50 a year by human analysts. Junior officers reportedly had only 20 seconds to approve strikes.
In the first two months of the war Israel struck 25,000 targets, four times the total in previous Gaza conflicts. Aviv Kohavi, a former IDF chief, claimed that the Gospel system had transformed operations. Whether every claim is correct is deeply disputed, but the direction is undeniable: AI is accelerating warfare faster than ethical frameworks can constrain it. And, under the barrage, civilians are suffering in ways that AI’s promise of targeted warfare did not envisage.
For Britain, all of this raises a dilemma. It cannot afford to fall behind. Allies will expect it to contribute to joint operations with AI-enabled systems. Adversaries will not hold back out of respect for ethics. Yet Britain’s armed forces are small, and their legitimacy rests heavily on compliance with international law. A scandal involving an AI-driven strike would be disastrous. Transparency would help reassure Parliament and the public. Diversifying suppliers would reduce reliance on American firms and give smaller British innovators a chance. And governance frameworks must move beyond platitudes. They must include audit trails, clear chains of accountability, and mechanisms to halt deployments when systems malfunction.
The SDR has made AI the emblem of Britain’s military modernisation. But ambition is not the same as achievement. For now, the MoD’s projects remain fragmented, its suppliers disproportionately foreign, and its oversight patchy. AI is used mostly for support, not strike. Yet the pressure to move from augmentation to autonomy grows with each passing conflict.
Britain insists it is preparing for a “new way of war”. The question is whether it will shape that way itself and do so with the ethics of warfare at the heart of its revolution, or whether, as so often in the past, the direction will be set by others.
Did you find this story interesting? Please support AOAV's work and donate.
Donate