The last couple of weeks of geopolitical events surrounding the war in Ukraine, and the US’s retreat from the world stage, have crystallised a truth that many in European tech have been repeating for some years now: the continent needs to bolster its own militaries and defence capabilities if it is to safeguard its democracies, and it needs to invest in new technologies to do so.
Yesterday Commission President Ursula Von Der Lyen announced an €800bn funding plan to rearm Europe, but if we want our defence sector to be fit for purpose on the modern, AI-powered battlefield, we need to urgently update the slow process of military procurement. This has been broadly true for many years, but in today’s AI software economy, where by far the greatest innovation is happening in the private sector and at unprecedented speed, the need for faster defence tech adoption has never been more pressing.
At the same time, the industry must ensure careful development, testing and deployment of this technology, to ensure it’s investing in battlefield systems we can trust.
In a world of autonomous drones, robots and AI-driven geospatial intelligence we need to develop standardised specifications for how we make autonomous systems ethical, robust and safe. That way innovators can get on with innovating, without having to second guess what kinds of assurances the buyers of their technology might be looking for.
Explainability
History offers us valuable lessons on how to accelerate technological adoption through standardisation.
Take the telecoms industry. By establishing international specifications for building, testing and deploying 3G, 4G and 5G networks — via frameworks like TEVV (Test, Evaluation, Verification, and Validation) — innovators and providers rolled out these foundational technologies at rapid scale and speed. Today, we don’t have such a well developed and standardised set of specifications for AI systems.
If Europe is serious about integrating AI into its militaries, we need a TEVV 2.0 tailored for AI. Central to this new set of specifications must be ways to develop and test AI systems that are as explainable as possible.
Technologies like autonomous drone swarms operating on the battlefield, each unit armed with a rifle, can’t be a black box where their actions can’t be scrutinised and understood by commanders on the ground. We must be able to evaluate and test how these technologies make life-and-death decisions.
Building battlefield-ready machines
Another sector that’s hugely benefited from these kinds of specifications is the automotive industry. Rules around safety features like seatbelts and airbags provided clear guidelines for manufacturers, freeing them to focus on advancing other aspects of vehicle design.
This blueprint approach saved time and ensured reliability. Now, consider bipedal robots — originally developed for civilian use cases like logistics or healthcare — which are increasingly likely to appear on future battlefields, and will require serious adaptation to be robust and combat-ready.
This will demand specifications around adversarial training, using simulations to prepare robots for encounters with enemy humans or rival autonomous systems. It also involves hardware considerations: Is the robot’s onboard GPU powerful enough to process vast streams of real-time data? Does its battery life support extended missions and a safe return to base?
Standardised guidelines in these areas would streamline development, ensuring that dual-use technologies transition seamlessly from civilian labs to the frontlines.
Ensuring fairness in high-stakes decisions
A third critical area for AI specification is fairness and bias mitigation. We’ve seen the importance of this in civilian contexts — such as in HR recruitment, where biased algorithms can unfairly exclude candidates — but the stakes are even higher in defence, where life and death decisions are being made.
One example would be the growing significance of geospatial data informing military intelligence.
The growing satellite market, which accounts for more than 70% of a space industry that’s expected to be worth $1.8tn by 2035, will continue to generate more and more data for both civilian and military use cases. AI will play a pivotal role in interpreting this geospatial intelligence.
Without designing systems that can understand their specific domain and context to test for unwanted bias, these systems risk skewing outcomes based on gender, race or other identity markers — decisions that could determine targets, troop movements or resource allocation. Establishing clear specifications for bias detection and correction is not just an ethical imperative, it’s a strategic one, ensuring that Europe’s AI-driven defence capabilities are both equitable and effective.
A race we can’t afford to lose
If Europe wants to keep pace — and protect its democracies — we must grease the wheels of defence procurement.
This applies equally to militaries sourcing from the private sector and to large primes integrating tech from smaller, nimbler innovators.
Both would benefit from clear metrics benchmarks, as this kind of framework would significantly accelerate acquisition and integration of AI software into sensitive environments, while reducing engineering time to release, deploy and operate AI systems in the field. If an AI system meets specification, it reduces the burdensome case-by-case due diligence needed to determine trust in it.
The urgency to establish standardised specifications for developing, testing and deploying AI in dual-use technologies is underlined by the fact that adversaries are already using advanced AI systems. We are fast approaching a world where the most powerful nations replace human soldiers and pilots on the frontlines with autonomous weapons.
Europe cannot afford to lag in this race, if we do we’ll be pitting our troops against AI-powered adversaries that literally never miss.
