Pioneering AI in defence: innovation, ethics, and the road ahead

Operating AI: trust above all

When deploying AI into decision making, the biggest challenge isn’t the algorithm, it’s trust. In high tempo, high risk scenarios, AI is only useful if operators believe it will behave predictably under pressure. That trust is earned through realism and rigour.

The foundations start with data availability and system integration. AI needs timely, relevant information and must be embedded seamlessly into operational workflows. Even with those pieces in place, people still need to see the system perform reliably in realistic conditions. Human-machine teaming should simplify the decision cycle, not add friction.

Ensuring reliability and ethical use requires strong assurance frameworks – clear guardrails, robust validation, and maintained human control. But we also have to avoid creating so much process that innovation slows to a crawl. The balance lies in enabling rapid experimentation while upholding the safeguards that defence and security demand.

Generative AI: connecting intelligence across the ecosystem

Generative AI is one of the most transformative technologies I’ve worked with. Its potential to optimise complex systems across defence is enormous, especially when you look at programmes end‑to‑end.

Generative models are already helping us capture knowledge, retrieve data quickly, and automate elements of software and product development. When combined with simulation and wargaming, for example, we have the opportunity to explore how systems might behave under different conditions long before they reach the field.

But the real power of generative AI comes from connecting these capabilities across the entire ecosystem. It can integrate data flows, support decision‑making from collection through to analysis, and enable a more coherent control of autonomous systems. In manufacturing, we’re seeing digital twins and AI‑driven optimisation are beginning to streamline processes and improve resilience.

The opportunity is huge, but so are the risks. Without robust safeguards, powerful point solutions can introduce new vulnerabilities. Assurance remains essential. We can innovate quickly, but we must do it responsibly.

Scaling AI infrastructure: secure by design

Developing AI infrastructure at enterprise scale in defence is all about making deliberate choices. We need performance, resilience, and security – and we need them across cloud, on premise, and edge environments.
AI at scale isn’t just about compute. It’s about supporting the full lifecycle: data access, model training, deployment, monitoring, and continuous improvement. Defence systems are inherently distributed, which means drones, sensors, vehicles, and ships all need to run performant models on constrained hardware, often in disconnected environments.

Balancing scalability with robust security means making trade offs. Cloud environments offer elasticity but require strong guardrails and data handling policies. On premise environments provide tighter control but need careful planning to scale. At the edge, the challenge is sharper still: delivering capable models that operate reliably without increasing operational risk.

Across all of this, assurance is the constant. You can scale aggressively, but without rigorous safeguards, you risk introducing vulnerabilities into mission critical systems.

AI driven cybersecurity: staying ahead of adversarial threats

Cybersecurity is increasingly an AI driven arms race. Offensive and defensive models evolve in parallel, and adversarial AI is becoming a very real threat. Addressing it starts with rigorous testing; challenging our models against adversarial techniques to understand how they could fail and how to strengthen them.

Provenance is equally critical. Knowing where every component, dataset, and model comes from allows us to react quickly when vulnerabilities emerge. There’s always a tension between assurance and speed of deployment, but in national security, agility can never come at the expense of safety.

AI’s role in safeguarding national security will only grow. It scales analysis across vast data streams, handles repetitive monitoring, and accelerates intelligence triage. It frees people to focus on higher order judgment. In that sense, AI becomes a genuine force multiplier when it is deployed responsibly.
 

Ethical AI: evolving with the mission

For me, responsible AI adoption starts with strong standards and a commitment to assurance throughout the lifecycle. We already have solid frameworks in the UK from MOD guidance to wider government and cybersecurity standards, and we see these internationally too. I advocate building on these rather than reinventing them.

Human control remains a core principle. Operators must be trained, empowered to override AI outputs, and supported by systems designed to prevent errors from escalating into harm. But ethical adoption can’t be static. It must evolve with the threat landscape and the tempo of operations.
 

Looking ahead

As I look to the future of AI in defence, one thing is clear: innovation and responsibility are not opposing forces. They are the foundations on which effective, trustworthy AI must be built. The challenge, and the opportunity, is to harness AI’s full potential while ensuring every capability is robust, assured, and aligned with the mission.