The killing machines taking over men in Ukraine

In June, Ukraine reportedly struck many targets in Mariupol connecting Russia to Crimea, using autonomous Hornet drones. These fully artificial intelligence-powered drones were programmed to lock on to targets, strike with precision, and cut off critical supply routes – all without human intervention.

Military planners facing a manpower crunch see a breakthrough being birthed. Defence planners, however, should awaken to this fresh warning.

In many ways, the ethical questions involved are not new. But the scale at which hundreds of drone manufacturers can churn out tens of thousands of small, cheap unmanned systems that are taking on leading roles in the conflict makes answering them imperative. 

Such autonomous systems are the key reason why a 40km chasm now separates the two front lines in eastern Ukraine, in which any moving person or vehicle is immediately spotted and attacked from above.

Innovations from the front line

Ukraine’s most important innovation is not the drone itself, but the software that powers it. The hardware assembled in Ukraine is not particularly ground-breaking. Early models mostly consisted of Chinese parts, augmented by 3D-printed materials, along with whatever scraps and improvised components soldiers can gather. 

The real breakthrough is the software that allows drones to navigate, identify targets and continue operating, even when communications are disrupted. As a result, military innovation cycles have compressed dramatically. New iterations can emerge in days rather than years.

Ukraine has been a living lab and test bed for new war-fighting concepts. There, more than 450 companies are constantly developing new tools, each a slight iteration of the last. 

This shift is transforming not just the battlefield, but also the business of defence. For decades, major contractors like Lockheed Martin, Raytheon and Boeing – the “neoprimes” – dominated military procurement deals through large, highly integrated weapons programmes that often took years or decades to research and develop, and more time to build on the factory floor.  

Today, a new generation of defence tech firms – Palantir, Anduril and more named after Lord Of The Rings references – are betting that software, autonomy and rapid turnaround will matter more than competing on hardware. Anduril’s Roadrunner drone interceptor, which can make autonomous decisions from sensor data and costs “in the low hundreds of thousands”, reportedly went from a sketch to operational validation and production in under two years. Anduril’s Australian arm took just three years from concept to deployment of its Ghost Shark underwater drones, which work from the same autonomous software. 

Even the established giants are adapting and pivoting. Lockheed Martin recently unveiled an unmanned naval drone system to compete with this new distributed manufacturing economy. The Lamprey multi-mission autonomous undersea vehicle can perform covert reconnaissance, quietly latch onto warships and deploy missiles and torpedoes.

Shifting procurement systems

Governments and military planners are responding accordingly, and redesigning defence procurement systems to accelerate acquisitions and encourage smaller firms to compete alongside traditional contractors. In 2023, the US Department of Defense announced the Replicator Initiative, designed to move the Pentagon’s famously slow rounds of bids and contractors towards prioritising real-time innovation, a defence strategy with a shortened deployment cycle and the scaling up of new defence tech. 

Start-ups no longer need to build entire defence platforms in order to participate in the industry. New companies jumping in may not themselves have the relationships or necessary security clearances to sell directly to the Pentagon, but they can work through larger firms like Palantir, which integrates the data infrastructure and sells it on. Drones and rockets could be replicated faster but the central challenge in building an autonomous tech stack remains.

The Trump administration, to its credit, has spent significant effort cutting down red tape to allow smaller firms to compete for contracts. Earlier in June, the US Air Force used novel contracting processes to separate hardware and software development for next-generation autonomous aircraft. They’re separating hardware and software development, so standard defence contractors like General Atomics and Lockheed Martin will develop the mainframe, while neoprimes like Shield AI and Anduril develop the autonomous software. 

This separation allows for much faster speed, but awarding contracts and seeing machines deployed on the battlefield are two different things that have yet to be connected. There are also concerns about oversight, which is more robust under normal contracts. 

Autonomy on the battlefield is quickly emerging with law and anti-proliferation work lagging far behind. A UN panel of experts found that a Turkish drone attacked retreating combatants without any human making a decision. AI systems used by the Israeli army have marked tens of thousands of Palestinians as targets without human oversight. And now in Ukraine, lethal drones are being deployed directly. 

The White House’s proposed US$1.5 trillion (S$1.9 trillion) defence budget for the next fiscal year shows the scale of resources now being directed towards these systems. 

Yet, simply spending more money on autonomous systems may miss the deeper lesson of Ukraine. Ukraine’s advantage lies not only in technological ingenuity, but also in organisational flexibility: engineers, front-line operators and manufacturers modifying systems quickly, testing them under combat conditions, and distributing successful adaptations throughout the network. 

Wartime necessity has created a culture of continuous experimentation that few peacetime bureaucracies can easily replicate. Hardware also can often be copied or manufactured at scale. But adaptive software, particularly systems capable of operating in contested environments where signals are jammed and conditions change rapidly, is much harder to reproduce in peacetime conditions.

International implications

Regardless, the implications far extend beyond Ukraine. As autonomous systems proliferate, the question is no longer whether they will become part of future warfare but how they should be governed. 

If such autonomous systems become the future of modern warfare, we must ask how they, like all weapons systems before them, should be regulated under international conventions. Technology being deployed on the battlefield in Ukraine is a test of both technology and of human capacity. 

It is tempting to think of nuclear weapons as the ready analogy. However, the investment required to engineer and build them means that only a handful of states can afford to acquire them. While their proliferation poses an existential risk at the use of even a single one, their spread is limited by the resources they require. 

Autonomous weapons are different. Their component parts are widely distributed, largely inexpensive and increasingly commercial in origin. A better comparison would be landmines or cluster munitions – cheap, effective and widely available. The treaty that limits their use was never signed by the major powers; international campaigns succeeded in stigmatising their use and reducing their legitimacy. 

Whether a similar approach can work for autonomous weapons remains uncertain, especially if they are increasingly viewed as essential to national defence. The battlefield in Ukraine suggests the future of conflict will reward investment in autonomous weapons. That creates powerful incentives for adoption and weak incentives for restraint and arms control.

Effective regulation will require countries to agree on what constitutes an autonomous weapon, understand the risks involved and recognise a shared interest in limiting their use. Yet at present, countries cannot even reach agreement on basic definitions. 

If anything, the war in Ukraine is not only birthing new military innovations but also fresh ethical dilemmas.

Across many sectors, AI is advancing faster than most institutions can understand or regulate meaningfully. Whether international cooperation and rules can emerge before the proliferation of such autonomous machines in conflicts remains an open question.