The west must hurry to catch up with Ukraine on AI combat

How dangerous is artificial intelligence? That question is gripping Silicon Valley right now, in the wake of an attack by a “swarm” of OpenAI AI agents on the Hugging Face platform.

Last weekend, in an underground bunker near the front lines of the Russian-Ukrainian war where two dozen commanders were glued to computer screens, I found myself pondering the same existential issue.

I had travelled there with a bipartisan group of US veterans-turned-politicians from the “With Honor” organisation, who backed a congressional bill to support Ukraine that just passed. What I witnessed was striking — and sobering.

Ukraine has long used innovative technology to defend itself, be that Starlink hardware, drones or home-produced missiles like the FP-7 “Flamingo”. Now AI is taking this to a new level.

Delta, Ukraine’s AI-enabled battlefield management system, turns vast amounts of data into usable information, while AI is helping soldiers to counter and launch drone swarm attacks, and/or conduct the “last mile” of targeting in order to avoid electronic jamming. Ground drones are also operating, with a range of 30km or more, to fire weapons, carry bombs, deliver supplies or evacuate wounded soldiers.

The latter is particularly important since the Russians deliberately attack medical teams. The Ukrainian army runs a video game-style reward system for its soldiers that awards more “points” for robotic medical evacuations than tank hits. “Robots save lives,” a tattooed young nurse told me.

So are those AI systems now conducting entirely autonomous attacks? Artem Martynenko, a top commander, was asked that at the recent Yalta European Strategy forum in Kyiv. But he chose not to answer, perhaps to avoid inflaming western fears. And as a new FT visual investigation shows, it is increasingly hard to define how far humans are remaining “in the loop”.

Russia is copying Ukraine’s innovations, reportedly with Chinese support, and recently launched a drone attack in Zaporizhzhia that seemed to be fully autonomous (that is, capable of selecting its target and striking it without operator intervention). Ukraine is experimenting too, as the FT report shows. Its Saker Scout drone reconnaissance system is a case in point.

So what should politicians — and investors — outside Ukraine conclude from all this? One obvious point is that Nato in general, and America in particular, have an overwhelming self-interest in backing Ukraine, both for moral reasons and to understand the future of war.

Second, Nato and western governments must urgently overhaul military procurement systems, which have hitherto focused on ultra-expensive items like jets, designed and delivered in multiyear cycles. Big kit remains important. But the wars in Iran and Ukraine have shown that cheap drones can destroy costly weapons. Or, as one seasoned US military operative in Ukraine told me: “Americans have always assumed their big wallet will win wars. Now that is wrong.”

Military experts also say that Steve Feinberg, the former financier now serving as US deputy secretary of defence, understands this well, and is fighting to include innovative venture capital-backed start-ups in the US procurement process. The setbacks in Iran have helped him get more support for this, investors tell me. And Mark Rutte, Nato head, is backing procurement reforms, arguing this week that “what we are learning from Ukraine is helping us strengthen our own defences”.

However, reform will be incredibly hard. Ukraine has made tech advances because its army uses tight feedback loops between “customers” (ie soldiers) and producers for speedy, iterative innovation. While that may be Silicon Valley’s approach, it is diametrically opposed to current western procurement models.

Third, western leaders must move on from a cold war mentality focused on a binary fight between superpowers. Yes, America and China are engaged in a geoeconomic fight for global hegemony. But in the military sphere, tech innovations are now making it easier than ever for minnows to hurt giants.

Fourth, western militaries must treat the Ukrainians as partners in order to learn about their AI-backed innovations in the skies, sunflower fields and underground. A recent Ukraine-UK deal around “Avengers” AI training data is a step in the right direction; hard-nosed realism is needed.

Of course, most western voters are (rightly) terrified about AI combat. And Anthropic, the US tech giant, has refused to let the Pentagon use its models without a human somewhere in the loop. Ukraine, by contrast, is less constrained because it is (still) outside Nato.

But AI pacifism is not a realistic stance for Nato when Russia, Iran and China are experimenting with AI-enabled robots. Western governments must work with Ukraine, devise basic ethical rules for the technology, and recognise that the AI military genie cannot be put back in the bottle — whatever we might wish.

[email protected]