The UK’s National Cyber Security Centre said in its Annual Review 2026 on Nov. 3 that AI is reshaping cyber risk mostly by making familiar operations faster, cheaper and easier to scale—not by replacing attackers with fully autonomous systems. For companies and critical-infrastructure operators, the immediate consequence is practical: less time to find exposed assets, patch weaknesses, spot phishing, and contain breaches before they spill into operational disruption.
That distinction matters because it answers the question executives should be asking. If AI mainly accelerates existing attack chains, what should organisations fund now? The NCSC’s own logic points away from speculative “robot hacker” products and toward response speed: accurate asset inventories, patching based on exploitability, tight identity controls, segmentation around safety-critical systems, tested backups, and better visibility across suppliers and operational technology.
Faster attacks, not magic attacks
The NCSC says it is almost certain AI will act as a force multiplier across reconnaissance, social engineering, vulnerability discovery and exploit-development workflows. In plain business terms, attackers can search code and internet-exposed services faster, draft more convincing lures at scale, and iterate exploit code more quickly after a weakness becomes known. That does not change their goals, but it compresses the period between disclosure, discovery, exploitation and patching.
The review also puts limits on the more dramatic narrative. Human operators are likely to remain essential for directing, validating and adapting AI-generated output, especially as campaigns meet real-world defences and unexpected failures. The NCSC says the biggest near-term effect is acceleration of existing tactics, techniques and procedures, with the greatest gains going to skilled operators. It also notes that threat actors can lose control of agentic systems, which can hinder operations rather than make them unstoppable.
That makes patching and response time a board issue, not just a security-team metric. AI-assisted vulnerability discovery and exploit development are highly likely, in the NCSC’s view, to shorten defenders’ patch window. The important measure is no longer whether an organisation has a written patch policy, but how fast it can move from disclosure to triage, testing, deployment, detection and containment across its most exposed assets.
Why critical infrastructure feels the pressure first
The warning lands hardest in operational technology, where mundane digital weaknesses can spill into physical consequences. Power switching, water controls, dams, manufacturing lines, heating and refrigeration all rely on connected systems that were often designed for longevity, availability or ease of maintenance rather than modern internet exposure. When those systems are visible online, running old software or left with default credentials, AI does not need to invent a new class of attack to raise the risk; it only needs to help adversaries find and exploit old weaknesses faster.
That is why the operational-technology example highlighted by The Associated Press matters. NCSC Director Richard Horne pointed to internet-connected refrigeration systems in France and the UK. The incident was not described as an AI attack. It involved old technology and default passwords. But it is a vivid illustration of the review’s main point: the next disruptive event may look less like science fiction than a familiar control failure hit with more speed and scale.
The NCSC says China is highly likely to have both the intent and position to use AI against critical national infrastructure, and it judges that AI malware is highly likely to be used in a ransomware attack on UK critical national infrastructure. Those are strategic assessments, not a named timetable or a claim that every actor has the same capability. AP’s framing of a likely AI-related attack on UK critical infrastructure by 2028 captures the direction of travel, but the more actionable message is that response windows are narrowing now.
What to fund before “autonomous defense”
If budgets are finite, the review suggests a clear order of operations. First, know what is exposed. Internet-facing services, remote administration tools, supplier connections and operational-technology assets need a current inventory. Second, set patch windows by exploitability and business impact, not by a flat calendar. A critical internet-exposed flaw on a safety-related or revenue-critical system deserves a different clock than a low-risk internal issue.
Third, strip out easy privilege. Default credentials, unnecessary administrator access and poorly governed service accounts turn a faster intrusion into a wider compromise. Fourth, segment safety-critical and hard-to-replace systems so a breach in corporate IT does not automatically spill into plant operations, water controls or cold-chain equipment. Fifth, maintain isolated backups and actually test recovery. Backups that exist only on paper are not resilience.
The same logic applies to AI use inside the organisation. The review warns that shadow AI can expose sensitive information, which means governance matters on the defensive side too. Employees need clear rules about what data can be entered into third-party tools, where AI outputs can be used, and who is accountable when automation touches customer, operational or security information. For many organisations, especially smaller operators, hospitals, manufacturers, local authorities and suppliers, this may be as important as buying another detection product.
The business effect will not be evenly distributed. The NCSC says organisations with mature security programs and resources are likely to benefit more from defensive AI than those stuck with legacy systems, weak governance and limited staff. That points to a widening resilience gap. Large enterprises with telemetry, patch pipelines and practised incident-response teams may use AI to trim analyst workload and speed triage. Smaller or asset-heavy operators may instead feel the technology mainly as adversary leverage unless they invest in visibility, recovery and disciplined basics.
That is also the right way to read the broader economic context. AP reported that malicious cyberactivity costs UK organisations almost $20 billion a year and that Jaguar Land Rover suffered more than $260 million in direct losses from a 2025 cyberattack. Those figures do not measure AI-caused damage. They show how expensive conventional cyber failure already is before any acceleration effect is fully felt.
So the useful readiness test is not whether a company owns an “agentic defense” platform. It is whether leaders can answer a few harder questions with evidence: How complete is the inventory of internet-facing and OT assets? How quickly can the organisation patch the vulnerabilities most likely to be exploited? Where do default passwords or standing admin privileges still exist? How often are backups restored in testing? How long does it take to detect, contain and recover from a supplier-linked or OT-adjacent incident? In the NCSC’s warning, that is where the AI story becomes a business story.
By
By
By
By
