UK cyber agency sets a 2028 deadline

Britain's cyber defense establishment has put a date on one of the hardest questions in security: when artificial intelligence stops being a productivity tool for defenders and becomes a weapon aimed at national infrastructure. The answer, per the UK's National Cyber Security Centre, is sometime between now and 2028.

The NCSC, the cyber arm of signals intelligence agency GCHQ, published its annual review alongside a separate assessment of how AI could reshape the threat landscape over the next two years. In it, the agency states that AI-powered malware is "highly likely" to form part of a future ransomware attack on Britain's critical national infrastructure.

That warning arrives with a blunt message for organizational leadership from the agency's chief executive, who used the report's release to argue that security spending has been treated as optional for too long.

What the agency actually said

The NCSC's warning is framed as an assessment rather than a prediction of a specific incident. It did not describe in detail what form AI-enabled malware might take, and the report does not name a particular group, target, or timeline beyond the 2028 window.

What the agency did point to is the reuse of tools built for defense. According to the NCSC, open source AI security tools originally intended to help security researchers are being repurposed for malicious activity. That is not a hypothetical: researchers have already documented widespread abuse of ARTEX, an open source, Chinese-made AI red-teaming platform, in attacks on UK registrar 123 Reg, Microsoft 365, and others since its launch in July.

The concern, in other words, is less about a novel superweapon than about existing offensive-capable tooling spreading further and faster than the teams trying to defend against it. ARTEX was built to probe systems for weaknesses. In the wrong hands, that same probing capability becomes an attack surface.

The pessimistic scenario in detail

The NCSC's assessment sketches a worst-case 2028, one in which it envisages "a significant increase in the frequency and sophistication of cyberattacks worldwide." Wider adoption of AI tooling, the agency argues, would also make attribution of attacks harder — a problem that cuts across both criminal and state-linked activity.

The report states that China "has likely played a role" in supplying such tools in this scenario, "enabling them to hide their activities within the noise," per the assessment. That framing ties the technical question of AI-assisted attacks to the geopolitical one of who can be held responsible when an intrusion cannot be cleanly traced.

There is a further scenario in the assessment that moves beyond the human-directed model: an autonomous attacker operating outside its human operator's control. The NCSC says it "is a realistic possibility such an entity could self-replicate and conduct cyberattacks at will, with no apparent motivation or restraint."

Attacks of that kind, the agency suggests, would complicate geopolitical relationships for the same reason as any other poorly attributed intrusion — countries that order attacks can more easily deny involvement when the trail is obscured.

"For decades, we have collectively got away with doing tech on the cheap."

— Dr Richard Horne, CEO of the National Cyber Security Centre

A chief executive's message to boards

Horne's public framing of the review focused less on the technical detail than on organizational priorities. He called on organizations to "invest with urgency" in their security and resilience.

In the same remarks, he set out a broader ambition for how UK organizations should approach the problem, describing a country where leaders "actively choose to invest with urgency" and where "all organizations make security and resilience a fundamental requirement, not an afterthought." He added that the collective result "would be a source of enduring national strength for us all."

Horne positioned the NCSC's Cyber Essentials certification as a starting point rather than a destination, suggesting organizations should scale their resilience efforts in line with their size, complexity, and risk.

Defensive AI, already in the field

The assessment does not treat AI as purely an offensive problem. The NCSC acknowledges that AI is already changing vulnerability discovery and patching, at a time when each Patch Tuesday seems to arrive with record numbers of bugs.

The agency is also testing whether the same technology can help defenders keep pace. Project DROMOS began earlier this year with a four-week pilot analyzing government code repositories before expanding into selected operational environments. The Government Cyber Coordination Centre (GC3) launched the project with the AI Security Institute and other departments. According to the NCSC, the pilot uncovered hundreds of potential security issues, including critical vulnerabilities.

The review also reaffirms plans for Cyber Shield, an agentic cyber defense system first teased in July. It is expected to be delivered in stages over the next five years, with the NCSC providing defensive agents deployed by participating organizations with data intended to strengthen national resilience. The agency is running the scheme alongside the Department for Digital, Culture, Media and Sport (DCMS).

The funding behind the plans

The government has committed a series of sums to back the strategy. As reported last week, £60 million has been set aside to fund the Cyber Shield initiative, alongside an additional £70 million for a new National Exposure and Real-Time Vulnerability (NERV) Centre, whose job will be to identify and maintain an ongoing understanding of vulnerabilities across critical national infrastructure organizations and the wider economy.

A further £40 million will fund a National Agentic Security Unit responsible for responding to uncontrolled bots and other agentic mishaps, and £24 million has been assigned to strengthen public sector defenses.

An undisclosed sum will also help establish a new NCSC-managed AI Cyber Security Centre in Manchester to develop AI security expertise across government.

• £60 million to fund Cyber Shield
• £70 million for the National Exposure and Real-Time Vulnerability (NERV) Centre
• £40 million for a National Agentic Security Unit
• £24 million to strengthen public sector defenses
• 75 percent discount on Cyber Essentials certification for micro, small, and medium-sized businesses applying for their first certificate

Where the certification fits

The government also announced a 75 percent discount on Cyber Essentials certification for micro, small, and medium-sized businesses applying for their first certificate. The discount lowers the entry cost for organizations that want to demonstrate baseline security practices.

Horne's advice to treat Cyber Essentials as a starting point, but to scale resilience in line with organizational size, complexity, and risk, sits alongside that discount. The agency's position is that certification alone is not the destination for larger or more exposed organizations.

An unresolved question on automated defense

The NCSC acknowledged that automated defensive activity of the kind envisioned in Cyber Shield currently sits outside many organizations' risk tolerances. The agency therefore has a role to play in building trust in the system before adoption can go mainstream.

That admission matters for how quickly the defensive plans move from pilot to practice. A system that deploys autonomous agents inside participating organizations requires those organizations to accept a degree of machine-driven action in their environments — a posture the NCSC itself says is not yet the norm.

Why it matters

The practical takeaway for UK organizations is that the NCSC is now naming a date, not just a direction of travel. If AI-powered malware is assessed as highly likely to feature in a ransomware attack on critical national infrastructure within roughly two years, the window for basic blocking and tackling narrows accordingly.

The evidence the agency cites is already concrete: existing red-teaming tools being repurposed, and at least one such platform — ARTEX — documented in attacks on 123 Reg, Microsoft 365, and others since July. Organizations that have deferred investment in resilience, particularly smaller suppliers who may now qualify for the certification discount, face a choice between using this period to scale up or treating the assessment as someone else's problem.

Horne's remarks point to the same conclusion from a different angle. The suggestion is that the years ahead will reward organizations that treat security and resilience as a fundamental requirement, not something to be revisited after an incident.