NCSC adds an AI chapter to its research problem book, setting out seven research questions

The UK's National Cyber Security Centre (NCSC) has added a chapter on AI to its research problem book, setting out seven research questions where it says "significant technical, operational and assurance challenges remain". The NCSC announced the chapter in a blog post published on 20 October 2026 by Paul J, its Technical Director for Cyber AI Research, and Brad A, a Principal Data Science & AI Researcher.

What the problem book is

The NCSC describes the problem book as a way "to guide cyber security research towards the most critical security challenges that we have identified as significant barriers to improving cyber security". It says the book is "primarily intended for academia, but it may also be useful to private sector companies carrying out research or identifying innovation gaps".

The book's other chapters cover cross-cutting problems, hardware security and cyber-physical problems. The blog says: "we’ve re-written our NCSC research problem book to include an entirely new chapter" on AI in cyber security.

The seven problems

The chapter lists them as high-level research questions:

1. "How can organisations re-design cyber defence workflows to maximise the benefits of agentic AI?"
2. "How can AI agents reliably help defenders identify, prioritise and reduce their attack surface?"
3. "How can AI strengthen threat detection, triage and incident response?"
4. "How should autonomous or semi-autonomous cyber defence systems be evaluated and assured?"
5. "How can individual AI and machine learning systems be secured throughout their lifecycle?"
6. "How do we secure current and future AI ecosystems?"
7. "How can data-driven methods improve cyber security analysis, posture and resilience at national scale?"

The blog groups these into three themes: "the use of AI to strengthen cyber defence", "the security and assurance of AI systems themselves" and "the use of data-driven methods to improve cyber resilience at scale". Both pages name human-AI collaboration as a thread running through all of them. The chapter says research is needed "to understand which cyber defence tasks are suitable for different levels of autonomy, where human judgement remains essential, and how to maintain meaningful oversight".

Threat actors and assumptions

The chapter says: "we have evidence that capable AI tools are being used by threat actors." It says AI "lowers the barrier to entry for some offensive tasks and increases the speed and scale of others", and that tasks such as writing exploit code are "increasingly AI-assisted". The page does not name specific incidents.

On the defensive side, it says that, used well, agentic systems "may help defenders operate at a speed and scale that would be difficult to achieve solely with human teams". It also cites an earlier NCSC blog, 'One does not simply defend agentically', as pointing out that, unlike attackers, defenders must use AI "within organisational, operational and governance constraints".

The blog adds that "Many of the assumptions that underpin today’s cyber security best practice were developed before this kind of AI existed." It says concepts such as vulnerability management, incident response and risk management "will need to evolve", and that this "may require a re-design of processes from the ground up".

The chapter says finding answers is also important for the NCSC and DCMS's work on Cyber Shield, which it describes as "a national-scale, collaborative approach to agentic cyber defence".

What it is, and is not

The NCSC frames the chapter as a set of questions, not answers. It says: "This list is not exhaustive and instead aims to provide a guide to the types of research questions where academia, industry and government can make tangible progress over the next 5 years." The book's introduction says it is "intentionally not an exhaustive list of problems, but instead a set of focus points".

Who the NCSC wants to hear from

The blog says the NCSC welcomes "feedback on the challenges identified in the problem book, suggestions for additional research questions, and examples of ongoing or planned research that could help address them", from "across academia, industry and government".

Contributions go through the problem book's feedback form. The form asks for a research title, a summary of 250 words or fewer, which problem (AI1 to AI7) the work relates to, and whether the sender has a previous relationship with the NCSC.

Why it matters

On our reading, the chapter is a research agenda: it tells universities and companies which AI security questions the NCSC wants worked on over the next five years, and asks for examples of ongoing or planned work. Earlier this month we covered the government's plans for AI and cyber defence in UK will use frontier AI to find weaknesses in government systems before attackers do, AI minister says. For the NCSC's existing advice on phishing-resistant sign-in, see Why a one-time code did not stop a China-aligned phishing campaign, according to Proofpoint, and what does.