Lethal Autonomous Weapons

Killing By Code: New Briefing and Dataset on UK Military AI Programmes

19/05/202603/06/2026Chris Cole

Drone Wars UK is today publishing the first of a series of short briefings on the new military companies who are pushing the use of AI in warfare in the UK. Alongside this, we are publishing a list of 26 key MoD programmes developing the use of AI for warfighting.

The Datafication of War: Palantir, AI and Algorithmic Violence

The first of our ‘Killing by Code’ briefings, The Datafication of War, examines Palantir and its involvement in UK military programmes.

Military AI firms such as Palantir are exerting significant influence over domestic defence programmes. These for-profit corporations are increasingly playing a key role in military ecosystems, mobilising R&D and facilitating the transfer of civilian technology to military applications. This interdependency with the state, indicates the growing militarisation of digital technologies, as well as their importance as military contractors.

With algorithms, Large Language Models, and autonomous systems permeating throughout the military domain, modern warfare is becoming increasingly reliant on data sets. Military advantage need not only rely on industrial output and hardware but on high-quality data and software. Whether that is to train military AI models that assist in the planning, or the use of data to execute combat missions with AI-assistance, warfare in the age of AI is experiencing a datafication.

Palantir is integral to this story. The company has been embedded within US national security structures since its founding in 2003 and is one of the most opaque yet influential corporations today. Put simply, Palantir is a private company which builds software for the military, intelligence agencies and transnational corporations to understand masses of data (data analytics).

This short briefing explores Palantir and their algorithmic and data systems. It examines the company’s origin and leadership; their footprint in different institutions within the UK and elsewhere; its key software and data analytic platforms and ends by looking at their involvement with the Israeli and Ukrainian governments during times of profound violence and war.

With AI software becoming more embedded in military and civilian decision-making chains, (military) AI companies stand to wield a significant amount of domestic and international power. It is only right that these companies are subjected to proper scrutiny.

Key UK Military AI Development Programmes

Drone Wars UK is also today publishing a list of key UK MoD military AI development programmes, together with brief details and current state of each programme.

While there are almost certainly other classified programmes ongoing in this area, this is the first time, to our knowledge, that a list of such programmes has been published.

Programmes to integrate AI into UK warfighting include Project Asgard (see here) which aims to develop AI-enabled strike capability for the British Army; Project Startle and SyCoiEA for the Royal Navy and programmes to develop an ‘Intelligent Virtual Assistant’ for the RAF’s future combat aircraft (Tempest) as well as Autonomous Collaborative Platforms (ACP). In addition Project Spotter and Project Odyssey are being developed to support UK Defence Intelligence. Alongside these are cross-service projects including the Digital Targeting Web, Sapient, adopted for counter-drone systems and Project Castle, using AI for cyber warfare.

The MoD has refused to give any details about three other RAF programmes involving AI on our list – Deep Thought, Omnia and Organon, either to us or national media.

A joint Royal Navy/RAF series of trials to validate AI algorithms is named Wintermute and may be named after one of the AI’s in William Gibson’s famed Neuromancer novel.

Drone Wars UK will continue to scrutinise developments in this area as part of our work to challenge the push towards the development of autonomous weapon systems.

UK MoD awards 26 companies contracts to develop AI targeting system for UK armed forces

12/03/202619/05/2026Chris Cole

While public concern about the use of AI for war-fighting continues to grow, the UK is quietly pressing ahead with development of new AI-based military targeting systems.

In a little-noticed post in January, the MoD named a group of 26 companies who have been awarded a four year deal to develop what it calls “advanced digital decision-supporting capabilities” as part of the ASGARD programme.

AI integration into military targeting system is developing rapidly. Image: Shutterstock

The group includes specialised US military AI company Anduril and Germany-based Helsing, traditional military tech companies like QinetiQ and Leonardo and a host of smaller niche companies focused on the use of AI. A full list of companies is below.

ASGARD

First announced in October 2024, the MoD says ASGARD will “exploit AI and novel communications networks” to provide “rapid targeting and decision-support to personnel.” While militaries are keen to use AI to speed up decision making around lethal strikes, there are serious ethical and legal concerns about these developments.

Use of AI by Israel to develop targets for strikes by Israeli during its war on Gaza and more recently by the US for strikes on Iran indicates that these developments are rapidly outstripping political and legal debate about whether these systems should be deployed at all. This week an investigation by Airwars and the Independent newspaper revealed that the US had accepted that a civilian had been killed in a series of US strikes carried out in February 2024, which at the time, the US said had been carried out with the assistance of Project Maven, a US programme to integrate AI/machine learning into military operations.

While continuing to argue in public that the UK has ‘no intention of developing a fully autonomous weapon’ the MoD also states that when “incorporating AI within weapon systems… there must be context-appropriate human involvement in [systems] which identify, select and attack targets.” This is vague to the point of meaninglessness and is impossible to know how such a policy will operate in practise.

ASGARD-image-1200x800 — A mock HQ utilising ASGARD at MoD press briefing, July 2025. Crown Copyright.

Accelerating Digital Decisions

The 26 companies have been awarded contracts in relation to a tender notice published by the MoD in July 2025, seeking companies to take part in an ‘Open Framework’ (that is, an ongoing development work) to develop AI/Machine Learning software to support decision making in military targeting for the British Army. As the tender notice stated:

“This Open Framework will focus on the ‘Decide’ element of the target acquisition cycle (Sense-Decide-Effect); supporting ASGARD’s goal of reinventing, and transforming, how land forces deliver operational decision-support and decision-making software via the use of modern Artificial Intelligence / Machine Learning (AI/ML) technologies.”

The Framework contains five separate ‘lots’ and the winning companies may be focusing on one or more of the different lots covering different aspects of the work. While some messaging around this Framework indicates the total amount to be awarded is between £180m and £216m, other indications are that this is the amount available for each lot. The MoD has said that ASGARD has been “backed by more than £1 billion in funding.”

The lots are as follows:

Lot 1: Data Integration

Work under this lot covers “higher-level functions like data validation, cataloguing, and lineage tracking. It will form the backbone for delivering trusted datasets supporting critical operations.” The tender notes that “basic cloud storage and compute will be covered elsewhere.”

Lot 2: Accelerators

Work under this area seeks software “to enhance data-driven decision-making… The focus is on reducing time-to-insight and improving operational efficiency… This lot targets intelligent capabilities such as automated workflows, pre-trained models, and integration with operational systems.”

Lot 3: Applications

The tender notice states that this lot “addresses platforms and services enabling mission-critical software to operate efficiently and securely across the enterprise… Focus areas include fast delivery, scalability, and continuous innovation.”

Lot 4: Edge Storage and Compute

‘Edge computing’ in this context means that processing and analysis is done ‘locally’ i.e within the surveillance or weapon systems and that video or other electronic information is not transmitted to a central control. The idea is that the drone, for example, processes the information it has captured itself rather than transmit it over networks to a central base for processing there. The tender says this lot “focuses on edge computing and local storage for real-time, low-latency data processing… Emphasis is on supporting distributed environments with limited or intermittent connectivity. This lot is essential for scalable, autonomous operations at the edge.”

Lot 5: Services

The tender states that this lot “includes expert services to support technology adoption and integration across all other lots. Offerings may include technical training, architecture consulting, synthetic data support, and proof-of-concept development.”

AI: speed eroding oversight and accountability

As we have said before, the grave dangers of introducing AI into warfare and in particular for the use of force are well known. While arguments have been made for and against these systems for more than a decade, increasing we are moving from a theoretical, future possibility to the real world: here, now, today.

Advocates of ASGARD and similar systems argue that the ‘need’ for speed in targeting decisions means that the use of AI brings enormous benefits. But while computer algorithms can process data much faster than humans, speeding up targeting decisions significantly erodes human oversight and accountability and will inevitably mean more civilian casualties.

While some argue almost irrationally in the powers and benefits of AI, in the real world AI-enabled systems remain error prone and unreliable. AI is far from fallible and relies on training data which time and time again have led to serious mistakes through bias. Most armed conflicts do not take place in remote battlefields but in complex and complicated urban environments. Relying on AI to choose military targets in such a scenario is fraught with danger.

The Companies Involved:

Collateral Damage: Economics and ethics are casualties in the militarisation of AI

20/08/202519/05/2026Matthew Croft

The current government places a central emphasis on technology and innovation in its evolving national security strategy, and wider approach to governance. Labour proposes reviving a struggling British economy through investment in defence with artificial intelligence (AI) featuring as an important component. Starmer’s premiership seems to align several objectives: economic growth, defence industrial development and technological innovation.

Rachel Reeves and John Healey hold roundtable with military company bosses, in front of Reaper drone at RAF Waddington, Feb 2025. Image: MoD

Taken together, these suggest that the government is positioning AI primarily in the context of war and defence innovation. This not only risks undermining the government’s stated ambitions of stability and economic growth but a strategy that prioritises speed over scrutiny, to the neglect of important ethical concerns.

The private defence industry has been positioned as an important pillar of this strategy. Before the Strategic Defence Review (SDR) was published, Chancellor Rachel Reeves and Defence Secretary John Healey initiated a Defence and Economic Growth Task Force to drive UK growth through defence capabilities and production. Arms companies are no longer vital for the purposes of national security but now presented as engines of future prosperity. AI is central to this, consistently highlighted in government communications John Healy has explicitly acknowledged that AI will increasingly power the British military whilst Kier Starmer stated that AI ‘will drive incredible change’.

UK focusing AI on military applications

The AI Action Plan, released in January 2025, explicitly links AI to economic growth. Although this included references to ‘responsible use and leadership’, the government has now shifted emphasis on military applications at the expense of crucial policy areas. On the 4th of July, the Science and Technology Secretary Peter Kyle wrote to the Alan Turing Institute – Britain’s premier AI research organization – to refocus research on military applications of AI. The Institutes prior research agenda spanned environmental sustainability, health and national security; under this new directive priorities are fundamentally being narrowed.

BAE Systems Project Odyssey uses AI and VR to make training ‘more realistic’. Image: BAE Systems

Relatedly, the Industrial Strategy released by the government aims to ‘embolden’ the UK’s digital and technologies economy, with £500 million to be delivered through a sovereign AI unit – this however will be focused on ‘building capacity in the most strategically important areas’. Given Peter Kyles re-direction and the overwhelming emphasis the government has placed on AI’s productive capacity in war, it becomes clear that AI research in defence will be at the cost of socially beneficial research in the case of the Alan Turing Institute.

Take Britain’s bleak economic outlook: sluggish productivity; post-Brexit stagnation; strained public finances; mounting government debt repayments; surging costs of living and inflating house prices. There is little evidence to suggest that defence-led growth will yield impactful returns on this catalogue of challenges. No credible economist is going to advise, in the face of these challenges, that investing in defence and redirecting research on AI in the name of national security, is going to give a better return on investment.

Research conducted in America illustrates that investing in health, education, infrastructure and green investment is more likely to give better returns on individual income specifically and broadly, the country’s prospects. Similarly, Lord Blunkett (former minster under Blair) pointed out that without GDP growth, raising defence spending as a share of GDP may not increase the actual funding.

Concerning applications to health outcomes, in August the World Economic Forum reported AI’s striking potential in doubling accuracy in the examination of brains in stroke patients, detecting fractures often missed in overstretched departments and predicting diseases with high confidence. This is critical given the NHS’s persistent challenges: long waiting times, underfunding, regional inequality, staff shortages and bureaucratic inertia.

Health and economic growth are closely related: healthier individuals are more productive, children attend school more consistently, preventative care lowers long-term costs – fundamentally strong health systems add value to the economy and our lives . Yet health is just one example. We are in the embryonic stages of AI development, and by prioritising research on military applications over civilian ones with public value, the government risks undermining, not fuelling long-term economic growth.

Crucially, framing arms companies as a major engine of economic growth is wildly misleading and economically unfounded. Arms sales account for 0.004% of the treasury’s total revenue and the defence industrial base accounts for only 1% of UK economic output. This sector is highly monopolized and so the benefits of ‘growth’ are concentrated among a handful of dominant corporations. Even then, the profit generated will not be reinvested into the UK. The biggest arms company in the UK – BAE Systems – is essentially a joint US-UK company with most of its capital invested in the US with majority shareholders emanating from US investment companies like BlackRock.

Prioritising speed over scrutiny

Beyond the economics, this is part of a wider strategy that signals a growing dismissal of ethical concerns, prioritising speed over scrutiny. The SDR acknowledged that technology is outpacing regulatory frameworks, noting that ‘the UK’s competitors are unlikely to adhere to common ethical standards’. In April 2025, Matthew Clifford – AI advisor to the PM – has been quoted saying ‘speed is everything’. While the Ministry of Defence (in 2022) promised to take an ‘ambitious, safe and responsible’ approach to the development of military AI, the current emphasis on speed sidelines important ethical concerns in the rush for military-technological superiority.

Militarily, the SDR makes plans to invest in drones, autonomous systems and £1 billion for a ‘digital targeting web’. A key foundational principle of International Humanitarian Law is the protection of civilians and their distinction with military targets. An AI-enabled ‘digital targeting web’ – like the one proposed in the SDR – connects sensors and weapons enabling faster detection and killing of human life. These networks would be able to identify and suggest targets faster than humans ever could, leaving soldiers in the best case, minutes, and the worst case, seconds to decide whether the drone should kill.

Digital Warfare: US and UK forces at the Combined Air Operations Center (CAOC), Al Udeid Air Base, Qatar,

One notable example is the Maven Smart System, recently procured by NATO. According to the US Think Tank, the Centre of Security and Emerging Technology, the system makes possible small armies to make ‘1000 tactical decisions per hour’. Some legal scholars have pointed out that the prioritisation of speed, within AI-powered battleground technology, raises questions surrounding the preservation of meaningful human control and restraint in warfare. Israeli use of AI-powered automated targeting systems such as ‘Lavender’ during its assault and occupation of Gaza is illustrative of this point. Systems such as these have been highlighted as one of the factors behind the shockingly high civilian death toll there.

This problem is compounded by the recent research that has shown that new large language models are known to ‘hallucinate’ – producing outputs in error or made up. As these systems become embedded within military decision-making chains, the risk of escalation due to technical failure increase dramatically. A false signal, misread sensor or a corrupted database could lead to erroneous targeting, or unintended conflict escalation.

In sum, the UK’s current approach – predominantly framing AI’s utility though the lens of defence – risks squandering its broader social and economic potential. The redirection of public research institutes, the privileging of AI investment in military applications (or so-called ‘strategic areas’) and the emphasis on speed over scrutiny raises serious concerns. Ethically, the erosion of meaningful human control in battlefield decision-making, the risk of AI-driven conflict escalation and the disregard of international humanitarian principles points to a troubling trajectory. The UK risks drifting towards the ethical standards of Russia and Israel in its use of military AI. A government approach to AI grounded in human security (freedom from fear and want), not war is not only more ethical but far more likely to generate sustainable economic growth for the United Kingdom.

Matthew Croft is a postgraduate student at Kings College London studying Conflict, Security and Development with a particular interest on the ethics of national security and the politics of technology.

UK crossing the line as it implements use of AI for lethal targeting under Project Asgard

23/07/202519/05/2026Chris Cole

Despite grave ethical and legal concerns about the introduction of AI into decision making around the use of lethal force, the UK is rapidly pressing ahead with a number of programmes and projects to do so, with the British Army recently trialling a new AI-enabled targeting system called ASGARD as part of a NATO exercise in Estonia in May 2025.

A mock HQ utilising ASGARD at MoD briefing, July 2025. Crown Copyright 2025.

Last week, the Ministry of Defence (MoD) gave a briefing to selected media and industry ‘partners’ on Project ASGARD – which it describes as the UK’s programme to “double the lethality” of the British Army through the use of AI and other technology. ASGARD is not aimed at producing or procuring a particular piece of equipment but rather at developing a communications and decision-making network that uses AI and other technology to vastly increase the speed of undertaking lethal strikes.

ASGARD is part of a £1 billion ‘Digital Targeting Web’ designed to “connect sensors, shooters, and decision-makers” across the land, sea, air, and space domains. “This is the future of warfare,” Maria Eagle, Minister for Defence Procurement and Industry told the gathering.

According to one reporter present at the briefing, the prototype network “used AI-powered fire control software, low-latency tactical networks, and semi-autonomous target recommendation tools.”

Janes reported that through ASGARD, “any sensor”, whether it be an unmanned aircraft system (UAS), radar, or human eye, is enabled by AI to identify and prioritise targets and then suggest weapons for destroying them. “Before Asgard it might take hours or even days. Now it takes seconds or minutes to complete the digital targeting chain,” Sir Roly Walker, Head of the British Army told the gathering.

Drones used in conjunction with ASGARD

While the system currently has a ‘human in the loop’ officials suggested that this could change in future, with The I Paper reporting ‘the system is technically capable of running without human oversight and insiders did not rule out allowing the AI to operate independently if ethical and legal considerations changed.’

How it works

A British Army report after the media event suggested that “Asgard has introduced three new ways of fighting designed to find, strike and blunt enemy manoeuvre:

A dismounted data system for use at company group and below.
The introduction of the DART 250 One Way Effector. This enables the targeting of enemy infrastructure three times further than the current UK land based deep fires rockets.
A mission support network to accelerate what is called the digital targeting or ‘kill’ chain.

According to a detailed and useful write-up of the Estonia exercise, ASGARD uses existing equipment currently in service alongside new systems including Lattice command and control software from Anduril which provides a ‘mesh network’ for communications, as well as Altra and Altra Strike software from Helsing used to identify and ‘fingerprint’ targets. The report goes on:

“targets were passed to PRIISM which would conduct further development including legal review, collateral damage estimates, and weapon-to-target matching.”

Helsing’s HX-2 drone was also used during the exercise and is another indication that the UK is likely to acquire these one-way attack drones. DART 250, a UK manufactured jet-powered one-way attack drone with a range of 250 km that can fly at more than 400 km/h was also deployed as part of the exercise. The manufacturer says that it can fly accurately even when GPS signals are jammed and that it is fitted with seeker that enables it to home-in and destroy jamming equipment.

AI: speed eroding oversight and accountability

The grave dangers of introducing AI into warfare, and in particular for the use of force are by now well known. While arguments have been made for and against these systems for more than a decade, increasing we are moving from a theoretical, future possibility to the real world: here, now, today.

Systems like ASGARD may be able to locate tanks on an open plain in a well-controlled training exercise environment (see video above), the real world is very different. Most armed conflicts do not take place in remote battlefields but in complex and complicated urban environments. Relying on AI to choose military targets in such a scenario is fraught with danger.

Advocates of ASGARD and similar systems argue that the ‘need’ for speed in targeting decisions means that the use of AI brings enormous benefits. And it is undoubtedly true that algorithms can process data much faster than humans. But speeding up such targeting decisions significantly erodes human oversight and accountability. Humans in such circumstances are reduced to merely rubber-stamping the output of the machine.

Meanwhile, the Ministry of Defence confirmed that the next phase of ASGARD’s development has received government funding while at the UN, the UK continues to oppose the negotiation of a new legally binding instrument on autonomous weapons systems.

Autonomous Collaborative Platforms: The UK’s New Autonomous Drones

13/06/202419/05/2026Chris Cole

Following on from the MoD’s Defence Drone Strategy released in February (see our report here), the RAF has now published its ‘Autonomous Collaborative Platform Strategy’ as it works to develop, produce and deploy these new type of military drones.

The strategy defines Autonomous Collaborative Platform (ACP) as types of uncrewed systems (drones) “which demonstrate autonomous behaviour and are able to operate in collaborative manner with other assets.” The strategy argues that Reaper and the (soon-to-enter-service) Protector drones “are vulnerable in warfighting conflicts involving peer or near-peer adversary. Therefore, as a priority the RAF needs to go beyond RPAS [Remotely Piloted Air Systems] to develop ACP capabilities.”

The plan argues that “through increasing use of autonomy, remote mission operators (commanders /supervisors) will be able to command an increasing number of AV [drones] within each ACP system.”

Underpinning the development, is the notion that the “geopolitical climate demands that we move beyond the caution of the post-cold war world” and that therefore the RAF must “undertake activity in areas that are demanding, difficult or overtly hostile.” While the Strategy sets out a variety of tasks for these new drones, it makes clear that a key focus is on “overwhelming an adversary’s air defences.” ACP are therefore not a defensive system, but are designed from the outset to enable the UK to engage in attack.

Tiers for Fears

The strategy sets out three ‘Tiers’ of ACP based on their ability to survive in “high-risk” (i.e. defended) environments:

Tier 1 ae disposable drones, with life-cycle of one or very few missions;
Tier 2 are “attritable” (or “risk tolerant”) that is, expected to survive but losses are acceptable;
Tier 3 are drones which have high strategic value, which if lost would significantly affect how the RAF will fight.

Diagram from Autonomous Collaborative Platform Strategy

Echoing the words of the Chief of the Air Staff Sir Richard Knighton before the Defence Select Committee earlier this year, the document states that a Tier 1 ACP will be operational “by the end of 2024”, while Tier 2 systems will be part of RAF combat force by 2030. Read more →

Proceed in Harmony: The Government replies to the Lords on AI in Weapon Systems

28/02/202409/04/2024Peter Burt

Last December a select committee of the House of Lords published ‘Proceed with Caution’: a report setting out the findings of a year-long investigation into the use of artificial intelligence (AI) in weapon systems.

Members of the Lords committee were drawn entirely from the core of the UK’s political and security establishment, and their report was hardly radical in its conclusions. Nevertheless, their report made a number of useful recommendations and concluded that the risks from autonomous weapons are such that the government “must ensure that human control is consistently embedded at all stages of a system’s lifecycle, from design to deployment”. The Lords found that Ministry of Defence (MoD) claims to be “ambitious, safe, responsible” in its use of AI had “not lived up to reality”.

The government subsequently pledged to reply to the Lords report, and on 21 February published its formal response. Perhaps the best way of summarising the tone of the response is to quote from its concluding paragraph: ““Proceed with caution”, the overall message of this [Lords] report, mirrors the MoD’s approach to AI adoption.” There is little new in the government response and nothing in it will be of any surprise to observers and analysts of UK government policy on AI and autonomous technologies. The response merely outlines how the government intends to follow the course of action it had already planned to take, reiterating the substance of past policy statements such as the Defence Artificial Intelligence Strategy and puffing up recent MoD activity and achievements in the military AI field.

As might be imagined, the response takes a supportive approach to recommendations from the Lords which are aligned to its own agenda, such as developing high-quality data sets, improving MoD’s AI procurement arrangements, and undertaking research into potential future AI capabilities. On the positive side, it is encouraging to see that in some areas concerns over the risks and limitations of AI technologies are highlighted, for example in the need for review and rigorous testing of new systems. MoD acknowledges that rigorous testing would be required before an operator could be confident in an AI system’s use and effect, that current procedures, including the Article 36 weapons review process, will need to be adapted and updated, and that changes in operational environment may require weapon systems to be retested.

The response also reveals that the government is working on a Joint Service Publication covering all the armed forces to give more concrete directions and guidance on implementing MoD’s AI ethical principles. The document, ‘Dependable AI in Defence’, will set out the governance, accountabilities, processes and reporting mechanisms needed to translate ethical policies into tangible actions and procedures. Drone Wars UK and other civil society organisations have long called for MoD to formulate such guidance as a priority.

In some areas the MoD has relatively little power to meet the committee’s recommendations, such as in adjusting government pay scales to match market rates and attract qualified staff to work on MoD AI projects. Here the rejoinder is little more than flannel, mentioning that “a range of steps” are being taken “to make Defence AI an attractive and aspirational choice.”

In other respects the Lords have challenged MoD’s approach more substantially, and in such cases these challenges are rejected in the government response. This is so in relation to the Lords’ recommendation that the government should adopt a definition for autonomous weapons systems (AWS). The section of the response dealing with this point lays bare the fact that the government’s priority “is to maximise our military capability in the face of growing threats”. A rather unconvincing assertion that “the irresponsible and unethical behaviours and outcomes about which the Committee is rightly concerned are already prohibited under existing legal mechanisms” is followed by the real reason for the government’s opposition: “there is a strong tendency in the ongoing debate about autonomous weapons to assert that any official AWS definition should serve as the starting point for a new legal instrument prohibiting certain types of systems”. Any international treaty which would outlaw autonomous weapon systems “represents a threat to UK Defence interests” the government argues. The argument ends with a side-swipe at Russia and an attempt to shut down further debate by claiming that the debate is taking place “at the worst possible time, given Russia’s action in Ukraine and a general increase in bellicosity from potential adversaries.” This basically seems to be saying that in adopting a definition for autonomous weapon systems the UK would be making itself more vulnerable to Russian military action. Really? Read more →