A video game is a strange place to look for defence insight, and yet here we are. Battlefield 6 is set a handful of years from now, and almost everything in it already exists in a laboratory somewhere.

That is what makes it worth a look. EA DICE did not invent a future so much as assemble one out of published research, then make it playable. For anyone working in defence technology it is a fairly honest preview of what the next decade feels like from the inside.

The simulation is now close enough to matter

Volumetric lighting, ray-traced reflections and neural upscaling were built to look good, but the by-product is a physics-accurate world that can be used for training. Commercial rendering and the kind of work going on in the Defence Digital Foundry are converging on the same problem from opposite directions, and the game industry has far more money to spend on it.

The opposition does not follow a script

Enemy squads learn. They adapt to how you play, which comes out of reinforcement learning work including DeepMind’s collaboration with EA. The military version of this is DARPA’s ACE programme, where agents learn air combat by playing themselves several million times, and the ambition behind the UK Defence AI Strategy. Agents trained in synthetic worlds then get deployed in real mission systems, which is either reassuring or alarming depending on the day.

The spectrum is treated as terrain

Players jam, spoof and reconfigure networks mid-fight, and the in-game effects of sensor denial and information fog are recognisable from real cognitive electronic warfare research. The defensive answer is the same in both worlds: assume a node will be compromised and design so that it does not take the network with it, which is the whole argument behind zero trust.

Swarms behave like swarms

Drones scout, relay, and occasionally throw themselves away to jam a radar. That is not a game mechanic invented for drama; it is close to the logic being trialled in programmes such as RAF Mojave and DARPA’s Gremlins, where attrition is a design assumption rather than a failure.

The kill chain is the game loop

Satellite, drone and ground mesh links feed command nodes that turn observation into an effect in a few seconds. Strip out the scoreboard and it is the same compression of the observe-orient-decide-act loop that real decision-support programmes are chasing, with the same dependency on fusing imagery, signals and human reporting into a single recommendation.

Weather is physics, not scenery

Local convection affects drone stability and laser coherence, which is exactly the kind of modelling that Met Office research and Copernicus observation data support. Once the environment is modelled properly, a mission twin stops being a picture and starts being a prediction.

Where it gets uncomfortable

The game asks the player to confirm lethal decisions proposed by an autonomous teammate. As a mechanic it is barely noticeable. As a depiction of human-on-the-loop governance it is more or less what the Responsible AI Senior Officers’ Report is arguing for, and it raises the same question: whether a confirmation prompt is oversight or theatre.

None of this turns a shooter into a wargame. But a commercial studio has produced one of the more accurate public depictions of near-future conflict in existence, largely by reading the same open research the rest of us do. That is worth noticing, if only because it tells you how little of this future is secret.