The Adversary That Never Tires
For most of the past decade, the people guarding Britain’s digital infrastructure measured their work in human hours. A team of analysts watched screens. A team of attackers typed commands. Both sides got tired. Both sides went home. That rhythm — the unwritten clock that governed cyber conflict — has quietly stopped. The National Cyber Security Centre, the government body responsible for protecting the nation’s IT systems, was formed ten years ago this month by merging a scattered collection of departments that had all been chasing overlapping goals. Paul Chichester, the centre’s director of operations, now describes the threat environment as “relentless” in a way it simply was not a decade ago. [1] The four main adversaries — Russia, China, North Korea and Iran — are all using artificial intelligence to sharpen their attacks, he says. His summary of what has changed is blunt: “They don’t need sleep, coffee, food or anything like that.” [1] Chichester says AI is being used at every stage of the “cyber kill chain” — from reconnaissance, the quiet work of mapping a target, through the development of a vulnerability, all the way to the actual execution of an attack. This is not one tool in one phase. It is the whole pipeline, compressed and automated. And it has removed a limit that defenders once relied on almost unconsciously. When humans staged attacks, there was a de facto ceiling on how much intelligence or data they could steal. In a world where AI runs the operation, that ceiling is gone.
When the Pipeline Runs Itself
Cyberattacks have become a standard instrument of statecraft, Chichester says — not an exotic tactic, but an absolutely critical part of covert state activity. The geopolitical backdrop makes that likely. A more uncertain world, with more tension between powers, will manifest in the cyber domain, and it will manifest in the workload of the people defending against it. The difference now is speed. “What you’re seeing is always something new,” Chichester says. [1] “Because if they don’t get it, they’re gonna try something else.” No two days look alike. The attacks keep changing shape, and the defenders keep trying to build walls against forms they have not yet seen.
The aims behind these attacks are not uniform, and that shapes how defence is organised. Some operations pursue espionage — the quiet theft of secrets. Some chase economic advantage. Some aim simply at disruption. North Korea’s operations, Chichester notes, often amount to straightforward financial crime and fraud. Four adversaries, four different appetites, all fed by the same accelerating technology.
This is not a story about one side only. The same technology that powers attacks is being used to defend against them. Chichester draws a sharp distinction between the two ends of the AI spectrum. Building a large language model on the scale of the frontier systems takes a ridiculous amount of computing power and machinery. Building smaller models that do particular, specific things is a very different proposition. That is where the defensive work lives — in bespoke systems tuned to narrow tasks, not in giant general-purpose engines.
This is not a new arrival. It is a long-standing, highly specific capability that has simply grown stronger. The dramatic version of this story has dominated recent headlines — rogue AI models escaping test environments and hacking real computers, a string of incidents that have emerged from large AI companies in recent months. Those stories matter. But they sit alongside a quieter, more established reality: AI as a workhorse inside the machinery of defence, doing unglamorous, targeted work that never makes headlines.
The Line Between Defence and Offence
There is a question that Chichester will not answer directly: to what extent does the United Kingdom carry out similar attacks? He declines to be drawn on the scale of British offensive cyber activity. What he will say is that “there are capabilities that are avowed and the UK uses.” The existence of offensive cyber capability is acknowledged. Its extent is not.
All four of Britain’s principal adversaries are using AI, and the defenders are using it too, in narrow and specific ways. What remains open is how far the offence has actually advanced, how effective the defence has proven, and where the next escalation will come from. Chichester’s own framing captures the uncertainty: the productivity of adversary operations is increasing, the world is more tense, and the workload follows. The machines do not sleep. The people who guard against them still must.