AI Cyber Attacks to Become Routine by 2028: EU & UK Threat Report

The NCSC's 2028 forecast marks a change in how intelligence agencies expect hostile governments to conduct cyber operations. Rather than treating AI as an experimental addition to their capabilities, states are expected to incorporate it routinely into intelligence gathering, offensive planning and technical exploitation.

That distinction matters. Cyber attacks have traditionally required considerable human effort to identify suitable targets, examine their systems and develop ways to gain access. AI can shorten several of those stages.

Reconnaissance can become more extensive. Social engineering campaigns can be produced in greater volumes. Vulnerability discovery and exploit development can proceed more quickly, reducing the time available for defenders to respond.

The underlying methods are hardly new. Phishing, stolen credentials and exploitation of known software flaws have been used for years. What changes is the cost of deploying them repeatedly and across large numbers of targets.

For a sophisticated intelligence service, this creates opportunities to expand operations without increasing personnel at the same rate. For less capable criminals, it lowers some of the technical barriers that previously limited their activities.

The consequences are already visible in Britain's incident figures.

During the NCSC's reporting period, its incident management team received 1,831 reports of cyber attacks. Of those, 337 required direct operational support, while 80 were classified as nationally significant.

Those figures do not establish how many incidents involved AI. They do, however, indicate the scale of the security workload into which increasingly automated attacks are being introduced.

The 2028 date is therefore less a prediction of a technological breakthrough than a warning about adoption. Britain has a limited window in which to strengthen systems before AI becomes an ordinary part of hostile state operations.

China and Russia Pose Different Cyber Threats

Although the NCSC identifies both China and Russia as major threats, their approaches are not identical.

China is assessed as Britain's most technologically sophisticated and sustained cyber adversary. Its operations form part of a broader state-directed effort to acquire intelligence, sensitive information and large quantities of data.

Data collection has acquired additional strategic value with the development of AI. Information obtained through cyber espionage can support intelligence analysis while also contributing to the training and development of advanced models.

The distinction between economic espionage and technological competition is becoming increasingly difficult to maintain.

China-linked organisations, including those associated with i-Soon and Integrity Technology Group, feature in the wider picture of sustained data acquisition and cyber activity. British authorities and international partners have responded to such threats through coordinated advisories and sanctions.

Russia presents a different problem.

The NCSC describes it as the most audacious cyber threat, reflecting an apparent willingness to conduct aggressive operations against infrastructure whose disruption could have consequences far beyond the computers directly affected.

One example is the December 2025 attack targeting Poland's electricity grid, linked in the briefing to FSB Centre 16, a Russian intelligence subdivision also identified as military unit 71330.

The operation had the potential to interrupt electricity supplies to 500,000 people.

The significance lies in the target. An attack on an electricity network is not simply an attempt to steal information or extort an organisation. It can threaten essential services and impose costs on the wider population.

Russian cyber operations against Ukrainian power infrastructure have already placed digital attacks within the broader conduct of military conflict. The Polish incident illustrates why European governments cannot assume that such threats will remain confined to an active war zone.

For Britain, the Chinese and Russian challenges demand overlapping but distinct responses. Protecting sensitive data and intellectual property is not quite the same task as preventing a disruptive attack on an energy network.

Both depend, however, on eliminating weaknesses that attackers can exploit.

The Problem Is Not Always Sophisticated Malware

One of the more revealing aspects of the NCSC's assessment is the continued importance of ordinary security failures.

Despite the resources available to advanced state actors, they still frequently rely on known vulnerabilities, compromised credentials and phishing.

There is a practical explanation. Attackers have little incentive to develop an expensive new exploit when an existing weakness provides access.

AI strengthens that calculation.

An organisation running outdated software may previously have benefited from the time and effort required for an attacker to discover its exposure. Automated vulnerability discovery can reduce that protection.

Once a weakness becomes known, attackers may be able to identify vulnerable systems and attempt exploitation before administrators complete their updates.

The result is a growing problem for organisations carrying years of accumulated technical debt.

Older applications, complicated infrastructure and poorly maintained systems are difficult to secure quickly. Replacing them can be costly, while interruptions to essential services may be unacceptable.

This creates an uneven security landscape. Well-funded organisations can invest in automated monitoring, modern architecture and dedicated response teams. Smaller businesses and operators dependent on legacy technology may struggle to maintain even basic protections.

The gap is particularly concerning where commercial systems support public infrastructure.

An energy operator or transport network does not operate in isolation. Its suppliers, contractors and technology providers can introduce additional weaknesses. A failure outside the organisation's immediate control may still affect its ability to deliver essential services.

AI does not create these dependencies. It makes their weaknesses more attractive to exploit.

Britain's Cyber Defences Face an Economic Test

The British government estimates that malicious cyber activity costs the country £14.7 billion annually.

That figure places cyber security firmly within economic policy, rather than leaving it solely to intelligence agencies and corporate IT departments.

Disruption can spread through production networks and supply chains. Major automotive industry incidents have illustrated how operational interruptions can affect businesses beyond the original target and contribute to weaker economic performance.

For companies, the costs extend beyond restoring compromised systems. Lost production, damaged intellectual property and interrupted commercial relationships can persist after the immediate attack has ended.

The NCSC's findings also expose a difficult investment problem.

Modernising infrastructure requires expenditure today, while the benefits often consist of incidents that never happen. For businesses under financial pressure, delaying upgrades can appear rational until an attack exposes the cost of that decision.

The government is seeking to change those incentives.

Its planned Cyber Security and Resilience Bill forms part of a broader effort to strengthen the regulatory environment surrounding essential services and their digital security.

The Government Cyber Action Plan includes £210 million in investment, while the Cyber Assessment Framework Core Profile establishes baseline security expectations.

The direction is towards more consistent standards and greater accountability. The precise obligations facing individual private-sector operators will depend on the legislation and its implementation.

Regulation alone, however, cannot resolve the speed problem identified by the NCSC.

From Periodic Checks to Continuous Cyber Defence

Traditional cyber security arrangements often depend on scheduled assessments, manual reviews and periodic software updates.

Those methods become less effective when attackers can continuously search for weaknesses using automated tools.

Britain's response increasingly includes machine-assisted defence through initiatives such as Cyber Shield and Project DROMOS.

The underlying ambition is to make protection more continuous: detecting threats, identifying vulnerabilities and supporting defensive action at a pace closer to that of the attackers.

This does not remove the need for human oversight. Security teams must still assess risks, prioritise critical systems and decide how to respond when protection measures could disrupt operations.

Nor does automation eliminate the problem of outdated infrastructure.

A sophisticated detection system cannot compensate indefinitely for software that remains unsupported or for credentials that are poorly protected. Automated defence is most useful when built upon systems that can be maintained and updated reliably.

The commercial implications are considerable. Developers of automated threat detection, agentic defensive tools and secure-by-design technologies are likely to encounter growing demand as organisations seek to reduce their exposure.

But the investment required will not be distributed evenly.

Large institutions may be able to overhaul their technology and introduce continuous monitoring. Operators of older infrastructure face a harder choice between expensive modernisation and continued dependence on systems that become increasingly vulnerable.

The challenge extends across European borders. Electricity networks, transport systems and commercial supply chains connect organisations with different resources, security standards and technological capabilities.

An attacker needs only one usable route into a network.

For British policymakers, the uncomfortable prospect is that AI-assisted offensive operations could become routine before defensive modernisation reaches the organisations most exposed to disruption.

The decisive question is no longer whether China, Russia and other states will use AI in cyber operations. Britain's intelligence agencies expect that to be normal by 2028.