The National Cyber Security Centre has delivered its starkest assessment yet of the threat posed by artificial intelligence, warning that AI-powered malware is “highly likely” to be deployed in a ransomware attack against Britain’s critical national infrastructure within the next three years. The verdict, published in the agency’s annual review, pins the blame squarely on hostile state actors — primarily Russia and China — who are expected to integrate AI “routinely” into their offensive cyber arsenals. For a sector already reeling from a relentless barrage of criminal extortion, the intervention marks a definitive shift: the theoretical risk of autonomous hacking tools has become an operational certainty.
The state of play: ransomware goes enterprise
Ransomware has long since graduated from a nuisance to a national security crisis. The playbook is familiar: infiltrate, encrypt, extort. Gangs operating from the former Soviet sphere have turned it into an industrial-scale enterprise, crippling the Royal Mail, the outsourcing giant Capita, and a rolling list of local authorities and public bodies. But the NCSC’s latest intelligence suggests the tradecraft is about to undergo a step-change. Machine learning models are now being weaponised to automate the grunt work of intrusion — scanning attack surfaces, fingerprinting vulnerable configurations, and even generating polymorphic malicious code that evades signature-based defences.
The review makes clear that the barrier to entry is collapsing. Where once a sophisticated intrusion required a team of skilled operators, AI agents can now compress the reconnaissance and weaponisation phases into minutes. That compression favours the attacker. It means a single threat actor can run parallel campaigns against dozens of targets simultaneously, testing defences at a scale no human team could sustain.
Horne’s warning: the floodgates are opening
Richard Horne, the NCSC’s chief executive, will use a speech on Tuesday to drive the point home. “In the coming months, we face a real prospect of an unprecedented rise in the number of cyber incidents,” he is expected to say. His language is deliberate. This is not a prediction of a single catastrophic event, but a warning of volume — a sustained surge that threatens to overwhelm incident response capacity across both the public and private sectors.
Horne will argue that AI’s true danger lies in its ability to excavate vulnerabilities that have been “tolerated for far too long.” The industry got a taste of that capability last month when researchers used large language models to uncover a critical flaw in PHP, the scripting language underpinning roughly 70 per cent of the world’s websites. The vulnerability had sat dormant for years. An AI found it in hours.
Professor Alan Woodward, a cybersecurity specialist at Surrey University, put it bluntly: “AI can help people find these flaws and patch them, but if we don’t get the access the bad guys will beat them to it.” The race is asymmetric. Leading AI labs such as OpenAI and Anthropic have begun granting early access to their most powerful models to select defenders, but the distribution is patchy. Most organisations — especially smaller operators in the supply chain — are flying blind.
Below the threshold: hybrid warfare by another name
The review also pulls back the curtain on the daily grind of state-sponsored aggression. British military communications networks are, Horne will reveal, subjected to “persistent” and “sustained” attacks. Russia is named as a principal perpetrator, but the campaigns are calibrated. They “intentionally fall below the threshold of armed conflict” while escalating attempts at hybrid warfare. It is a strategy of erosion: degrade confidence, sow friction, test red lines — all without triggering Article 5.
The supermarket fridge incidents in Europe, where attackers raised temperatures in cold storage units above 50°C, serve as a grim proof of concept. They demonstrate that operational technology — the systems that control physical processes — is no longer air-gapped or obscure. It is reachable. And it is fragile.
“For decades we have collectively got away with tech on the cheap,” Horne will say. The bill is coming due.
Why it Matters
The NCSC’s timeline — 2028 — is not arbitrary. It aligns with the maturation curve of generative AI tooling and the projected proliferation of autonomous agent frameworks. By then, the distinction between a state-grade capability and a commodity criminal toolkit will have effectively vanished. The implication for the UK is unambiguous: resilience can no longer be treated as a compliance exercise. The government’s “Be Ready” campaign, urging households to stock water and food, is a tacit admission that the state cannot guarantee the continuity of essential services in the face of a determined, AI-augmented assault. For boards and ministers alike, the choice is no longer whether to invest in cyber defence, but whether they can afford the consequences of failing to do so.
