NCSC Annual Review 2026

Frontier AI systems mean tasks that once required specialist knowledge and skills can increasingly be accessed by a broader cross-section of individuals and organisations through smaller and cheaper and open models. The result is a democratisation of technological capabilities, with more people creating software, analysing data and automating tasks.

This democratisation has significant defensive potential. Organisations that previously lacked access to specialist cyber expertise can access sophisticated security capabilities, while skilled cyber security professionals can automate routine analysis and spend more time on judgement, research and the hardest problems.

But democratisation also enables malicious use, helping less capable attackers to adapt or produce malware, and automate the exploitation of vulnerabilities. Agentic systems can be given an objective, select tools, sequence actions, inspect results and adapt their approach with varying degrees of human oversight. The consequence is not new categories of cyber risk, but rather the sharp increase in the speed, scale and persistence of familiar risks. 

In cyber security, this matters because security controls have traditionally been built around human limitations. Security experts work at human speed, require rest, cannot inspect every event and operate within social and organisational constraints. Agents can run continuously, operate multiple tasks and act across many systems at once. A moderately capable system operating at scale can therefore be consequential, even without making wider conceptual leaps traditionally associated with expert-led cyber security development.

Over time, the same capability will support a correction. Technology vendors will use AI throughout the development lifecycle to identify flaws before release, generate and test fixes, and protect customers of managed services. Defenders will use agents to discover network vulnerability, triage alerts, map networks and rehearse response. The structural advantage will remain with defenders; they own or operate the environment, can instrument it, can change it and can define acceptable behaviour. But this advantage is not a given, and realising it requires clear direction, investment and disciplined adoption.

Equally important is ensuring organisations develop AI systems securely, whether those systems have been created from scratch or built on top of tools and services. As our world-leading guidance on secure AI development points out, organisations need to understand what data a model can access, what tools an agent can invoke, what instructions it trusts and how its actions are observed and constrained. This means security must be considered across the whole AI lifecycle: design, development, deployment, operation and maintenance.

As AI becomes embedded in connected devices that sense and act in the physical world - think robotics, autonomous vehicles, drones, connected medical devices and industrial systems - the consequences of compromise will also change. Where systems interface with the physical world, security priorities must include safety, reliability, availability and resilience, not just information protection. Cyber security increasingly becomes a societal issue rather than a purely technical one, as the attack surface expands to include trust in information, institutions and digital interactions, creating new challenges around fraud, misinformation and authentication.

An approach based on discovering and manually patching every flaw cannot keep pace with machine-speed vulnerability discovery. Automated remediation will be part of the answer, but automation applied to insecure foundations does not address the fundamental issues. The longer-term goal must be technology that prevents broad classes of vulnerability, and limits the consequence of those that remain.

This requires investment across the technology stack. Secure development practices can make vulnerability prevention part of normal engineering. Hardware-backed security, strong identity and trustworthy update mechanisms can provide dependable control points. This connects to NCSC’s work on cross domain technologies: as AI, cloud, operational technology and higher-assurance environments become more interconnected, we will need ways to move data and insight safely between differing trust domains.

The long-term goal is technology that is secure by design and resilient by default. But while these technological shifts gather pace, organisations must still take practical action to reduce risk today. The challenge for the next decade is combining stronger foundations with consistent implementation of proven cyber security practices. The NCSC’s Principles Based Assurance will help with this, moving assurance away from static compliance gained from a one-off assessment, towards confidence in security outcomes. It asks whether a product, system or service can make credible security claims, support those claims with evidence, and maintain that confidence as technology and threats change. That approach matters even more as AI accelerates vulnerability discovery and increases the pressure on organisations to patch and respond quickly.

Roles and skills will also change. AI can raise the productivity of experienced practitioners by allowing multiple lines of analysis, and a more strategic focus of effort. It can help those new to cyber security learn and broaden participation. However, if entry-level tasks are automated without redesigning training pathways, the profession may inadvertently weaken the pathways through which people acquire deep expertise. Organisations should preserve deliberate practice, mentoring and opportunities to work through problems rather than only review machine-generated answers.

Over the past decade the UK has built one of the world's most respected cyber security research ecosystems, bringing together government, academia and industry to address some of the hardest cyber security and technology challenges we face. This ecosystem has consistently delivered advances that improve our national security while also supporting economic growth, innovation and broader skills development.

At the heart of the NCSC's technical capability are our own experts, who combine operational experience with long-term research and technology leadership. Their work helps identify emerging risks, shape standards, influence global technology development and ensure that security considerations are built into new technologies from the outset.

Research matters because it delivers practical outcomes. Examples such as SilentGlass - a device that blocks anything unexpected or malicious between HDMI and Display Port connections and screens - demonstrate how long-term investment in cyber security research can translate into capabilities that directly improve national security and resilience. 

Within the UK research community, the NCSC Academic Centres of Excellence in Cyber Security Research (ACE-CSR) network brings together universities undertaking internationally recognised cyber security research across a wide range of disciplines, from secure systems and cryptography to human factors, artificial intelligence and cyber-physical resilience. Alongside this, the NCSC and UK Research and Innovation (UKRI) support a number of Cyber Security Research Institutes (RIs) that focus expertise on strategically important national challenges. Together these partnerships create a powerful national capability, provide pathways for new ideas to become deployable technologies, and help develop the next generation of cyber security professionals that the UK economy needs.

AI may dominate the immediate technology landscape, but it is not the only major transition. Over the past decade, our work on post-quantum cryptography (PQC) has ensured our most critical systems are resistant to the threat, and that we can provide world-leading guidance to help UK organisations migrate safely to PQC.  We were the first major authority to set out publicly our early thinking on PQC, in a white paper published just 2 months after the NCSC was formed in 2016, and have produced influential guidance ever since.