Why military autonomy should think like a brain, not a black box

The autonomous drone team had already found the target. It had worked out the safest route, selected the most appropriate platform for the strike and calculated the probability of success. Then it stopped. Not because it was confused, but because it knew the next decision wasn't its own to make.
That moment captures one of the biggest misconceptions about military autonomy. Success isn't measured by how often an autonomous system acts alone, but rather knowing exactly when it shouldn't.

Much of today's discussion around artificial intelligence assumes autonomy is a single capability – a sophisticated algorithm that observes the world, reaches a conclusion and acts. This is an attractive idea, but it bears little resemblance to how experienced military professionals make decisions.

A soldier reacting to incoming fire isn't using the same cognitive process as an intelligence analyst piecing together evidence over several hours. A mission commander planning tomorrow's operation isn't thinking like a pilot avoiding a collision. And none of them operate without clear rules, training and authority governing every decision they make. The human brain doesn't rely on one form of intelligence and neither should military autonomy.

Imagine a reconnaissance mission searching for mobile air-defence systems that threaten a brigade's advance. Using an autonomous drone team, an operator doesn't need to micromanage every platform. Instead, the operator defines the mission intent: identify threats within this area, recommend the safest opportunity to neutralise them and position assets in readiness to proceed. Behind that deceptively simple instruction, several very different forms of reasoning are already at work.

Some decisions need to happen almost instantly. Avoiding a mid-air collision or obeying an urgent human command require the autonomous equivalent of reflexes – simple, near-instant responses. Other decisions require tactical judgement. A software-defined radio detects an unfamiliar transmission. One platform quietly repositions to improve reception while another is tasked to provide visual confirmation. This requires high-speed reasoning, in the same way as an experienced operator would rapidly assess options and issue commands.

Above this sits deeper reasoning. The mission itself is continuously re-planned as circumstances change, balancing endurance, sensor coverage, communications resilience and the availability of multiple autonomous platforms. These are all continually optimised to achieve the commander's objectives.

And then there is the most important layer of all – governance. This isn't another AI making recommendations, but a deterministic layer that continually checks every proposed action against defined constraints: rules of engagement, human authorisation, safety policies, geofences and operational boundaries. It doesn't matter whether a decision originates from a rapid tactical response or a longer-term mission plan – every action is verified before it is executed.

This mirrors how the military chain of command has always worked. Every individual – from the most junior operator to a senior commander – acts within a defined set of permissions. They know what decisions they're authorised to make, which situations require them to pause and seek guidance, and which actions sit above their level of authority entirely. A human drone operator today works within clearly defined authority. They fly within strict guidelines that define what they can and can't do, and they know when a change in circumstances means they need to refer upward before proceeding. That framework of delegated authority and escalation isn't a constraint on effectiveness – it's what makes the system trustworthy.

When we talk about AI-driven autonomy, we're describing exactly the same principle applied to a machine. The autonomous system isn't granted open-ended freedom to act. It's given permission to operate within a clearly defined boundary – just as the human operator it works alongside is trusted to fly the drone, but isn't the one making mission-level decisions. The technology changes. The logic of delegated authority doesn't.

That's why the drone team paused – not because autonomy had failed, but because trust had succeeded.

As defence organisations around the world explore increasingly capable autonomous systems, our challenge is no longer building machines that can think. It is designing systems that think appropriately, understand the limits of their own authority and always leave the right decisions in human hands.

For organisations like BAE Systems, that challenge extends far beyond artificial intelligence itself. It draws on decades of operational experience across air, land, maritime, cyber and electromagnetic warfare, working alongside armed forces to understand not just how platforms perform, but how commanders make decisions under pressure. That same experience now informs our work on trusted and assured autonomy. It is this expertise that enables us to develop autonomous systems that deliver successful mission outcomes, within approaches that ensure the capability remains safe, accountable and aligned with operational and ethical expectations.