Defence AI: opportunity outpacing governance?
Jonathan Lee
Director of Cyber Strategy , TrendAI
The UK Ministry of Defence is moving quickly to embed large language models (LLMs) across its operations. From intelligence analysis to logistics and training, AI is already reshaping how decisions are made and how fast information can be processed. This is no longer experimental. It is active, scaled and increasingly embedded into core workflows, with the Defence AI Centre and Dstl driving adoption.
The use cases are practical and immediate. LLMs are being used to interpret complex outputs from wargaming simulations, process large volumes of reconnaissance data and surface critical insights faster than traditional methods. They are helping reduce cognitive overload by filtering intelligence into something usable in high-pressure environments. Elsewhere, secure internal chatbots are improving access to procurement policies and technical documentation, while AI-powered search is enabling personnel to query vast datasets using natural language. In training, multimodal models are connecting real-time and virtual environments to create more advanced simulations.
This is not marginal improvement. It is a shift in how defence operates, with AI moving from support function to something closer to operational backbone.
Alongside this progress, governance is evolving, but not at the same pace. Frameworks such as JSP 936 set out clear principles around human oversight, accountability and risk mitigation. There are also efforts to introduce secure access layers and controlled validation environments to test models before deployment. However, oversight remains fragmented and often tied to specific use cases rather than a unified approach. There is no single approved model framework, and responsibility is still distributed across different parts of the organisation.
That creates a growing gap between capability and control.
The challenge is compounded by the ecosystem supporting this shift. The MOD’s AI strategy relies on a wide network of partners across cloud, research and defence technology. This accelerates innovation, but it also introduces complexity. Each integration point expands the attack surface, and every model must be secured, validated and continuously monitored. At scale, maintaining visibility across this environment becomes increasingly difficult.
What is happening in UK defence reflects a wider reality. AI is no longer a future capability. It is becoming part of core infrastructure. The question is no longer whether it delivers value, but whether organisations can maintain control as adoption accelerates.
If governance, visibility and security do not evolve alongside deployment, the systems designed to create advantage risk introducing new forms of exposure. Defence may be leading the way, but this is a challenge that extends far beyond it.
