TL;DR
The National Cyber Security Centre published a blog post on 20 October introducing a rewritten version of its research problem book, which now carries a full chapter on AI in cyber security. The chapter lists seven research areas under three headings and asks researchers to help answer them. The NCSC wants feedback on the questions it has chosen, and examples of work already under way.
What the chapter covers
The NCSC says it rewrote the problem book to help focus innovation as AI changes the field. The new chapter groups seven areas into three themes:
- using AI to make cyber defence stronger
- securing and assuring AI systems themselves
- data-driven techniques for making resilience work at large scale
Running through all three, the authors say, is the question of how people and AI work together: where autonomy adds real value, where human judgement is still needed, and how oversight can survive when machines act at machine speed.
Why the NCSC is doing this
The post argues that newer generative and agentic AI systems, which can plan, use tools and act with growing independence, challenge assumptions that today’s security good practice was built on, since much of that thinking predates these systems. The authors expect work such as incident response, managing vulnerabilities and assessing risk to change, to cope with AI-powered systems and AI-powered attackers alike. Some processes, they suggest, may need redesigning from scratch.
The blog links the chapter to two earlier strands: a Five Eyes warning that AI is speeding up cyber threats, and the NCSC’s Cyber Shield initiative, which sets out how AI could strengthen defence. Turning that vision into practice, the authors write, depends on answering hard questions about design, assurance, security and trust.
The post is signed by Paul J, who directs technical work on cyber AI research at the NCSC, and Brad A, a principal data science and AI researcher on secondment through the Industry 100 scheme.
Looking forward
In our view, this is the NCSC telling universities and suppliers where it expects AI security research to be useful, following its operations director’s recent warning about hostile states using AI in attacks. For UK security firms and researchers, the chapter is a public map of the problems the national authority wants solved, and responding through the NCSC’s feedback form is a low-cost way to shape what comes next.