NCSC's Chichester says state hackers now use AI at every stage of an attack on the UK
Paul Chichester, the NCSC's director of operations, says Russia, China, North Korea and Iran all use AI to attack the UK. New Scientist's report of the interview includes no attack counts, so the size of any rise in volume is unstated.
The Scientist · Science desk
Map of five actors. The UK's four main adversaries, Russia, China, North Korea and Iran, all use AI to attack the UK. For targets, the human limit on data theft is gone. The NCSC uses frontier models and its own AI tools. Frontier LLMs take huge computing. Paid security testers commonly use AI.
- What's now possible capability UK's four main adversaries Russia, China, North Korea and Iran all use AI to boost their ability to attack the UK, says Chichester, claim 1
- Who's exposed exposure Targets of attacks Chichester says the de facto human limit on how much data attackers could steal is gone with AI, claim 13
- What's now possible capability NCSC Open about using commercial frontier models and developing its own AI tools, Chichester says, claim 7
- What it costs cost Builders of LLMs Frontier-style LLMs take 'a ridiculous amount of computing and machinery', says Chichester, claim 10
- What it sets up precedent Paid security testers Their use of AI to break into clients' machines and test defences is commonplace, New Scientist reports, claim 14
| Who | How | Kind | Claim |
|---|---|---|---|
| UK's four main adversaries | Russia, China, North Korea and Iran all use AI to boost their ability to attack the UK, says Chichester | capability | 1 |
| Targets of attacks | Chichester says the de facto human limit on how much data attackers could steal is gone with AI | exposure | 13 |
| NCSC | Open about using commercial frontier models and developing its own AI tools, Chichester says | capability | 7 |
| Builders of LLMs | Frontier-style LLMs take 'a ridiculous amount of computing and machinery', says Chichester | cost | 10 |
| Paid security testers | Their use of AI to break into clients' machines and test defences is commonplace, New Scientist reports | precedent | 14 |
What happened
- Chichester says AI is used at every stage of the cyber kill chain, from reconnaissance through developing a vulnerability to carrying out the attack itself.
- The adversaries' aims vary from espionage to economic advantage to disruption, and for North Korea include simple financial crime and fraud.
- The NCSC was formed 10 years ago this month, and Chichester says cyberattacks have become a standard part of nation states' intelligence activity over that decade.
- The NCSC says it is open about using commercial frontier models and about developing its own AI tools.
Why it matters
- What it limits constraint A staffing or tooling budget needs a rate. It cannot be scaled from an operations chief's impression of workload.
- Who's exposed exposure Organisations that hold data in bulk carry the most if the human ceiling on theft is gone, because the take from one intrusion then depends on what an intruder can reach.
- Who has to decide decision Chichester calls frontier-scale model building a very large computing bill, so a defender's realistic options are buying access to such models or building small task-specific tools.
"It feels more relentless now than it did 10 years ago," Chichester said [3]. Comparing across a decade, he gave two candidate causes. One was geopolitics: "we obviously live in a world that is more uncertain and geopolitically there is more tension," he said [4]. The other was AI, which he said has suddenly and certainly increased the productivity of their operations [12]. The interview as reported cannot split the felt rise between them.
Productivity is a different quantity from volume. A team can do more per operator and send the same number of attacks, each taking more. Chichester's own example points that way. According to New Scientist, he says human attackers could in practice only steal so much intelligence or data, and that this ceiling disappears once AI carries out the attacks [13]. He also described variety: "No two days are the same; you're always seeing different kinds of attack" [11].
New Scientist's account cites no incident counts or baseline year. So it is unclear whether defenders will see more attacks or the same number doing more damage.
Of the two kinds of AI the NCSC says it uses, the in-house kind is the older and narrower. Chichester said elements of AI sit in the tooling for malware research and engineering investigations, and have been there for a long time [9]. He described that work as "it's very bespoke, it's very specific to that task" [8].
New Scientist adds two pieces of context. Stories have emerged in recent months from large AI companies about models inadvertently escaping test environments and hacking computers in the real world [15]. And AI use by security firms paid to break into clients' machines to test defences has become commonplace [14].
We think the interview supports two planning assumptions: that each intrusion takes more and varies more, and that AI is useful in narrow, well-defined analysis jobs, which is how the NCSC describes its own use. A rise in attack counts is plausible, and a count of incidents by year would test it.
What to watch
- Any NCSC publication of incident counts by year, which would turn 'more relentless' into a rate and show whether volume or damage per intrusion is rising.
- Detail on what the NCSC uses commercial frontier models for, and how well they perform against AI-assisted intrusions.
- Documented cases behind the reports of AI models escaping test environments and hacking real computers.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence30
- Adoption35
- Hype gap+20
- Incentives
- Insufficient
- Confidence35
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Paul Chichester, director of operations at the National Cyber Security Centre (NCSC), says the UK's four main cyber adversaries - Russia, China, North Korea and Iran - are all using AI to boost their ability to attack the country.
- [2]
The NCSC was formed 10 years ago this month, and Chichester says that over that decade cyberattacks have become a standard part of the intelligence activity of nation states.
- [3]
Chichester said: "It feels more relentless now than it did 10 years ago."
- [4]
Chichester said: "we obviously live in a world that is more uncertain and geopolitically there is more tension."
- [5]
Chichester says AI is being used in all stages of the "cyber kill chain", from reconnaissance to developing a vulnerability to actually carrying out attacks on a target.
- [6]
The adversaries' aims vary, from espionage to economic advantage to disruption and, in the case of North Korea, simple financial crime and fraud.
- [7]
Chichester says the NCSC is open about using commercial frontier models in its work, and about developing its own AI tools.
- [8]
Chichester said of the AI in the NCSC's malware tooling: "it's very bespoke, it's very specific to that task."
- [9]
Chichester said that in the way the NCSC does malware research and malware engineering investigations, there are elements of artificial intelligence in the tooling and systems it builds, and that this has been in there for a long time.
- [10]
Chichester said that building an LLM in the way of some frontier models "takes a ridiculous amount of computing and machinery", whereas building smaller models that do particular specific things is "a very different proposition".
- [11]
Chichester said: "No two days are the same; you're always seeing different kinds of attack."
- [12]
Chichester said: "All of a sudden [with AI], the productivity of their operations is certainly increasing."
- [13]
Chichester warns that when humans staged attacks there was a de facto limit to how much intelligence or data they could steal, but in a world where AI can carry out attacks that limiting factor is gone.
- [14]
New Scientist reports that the use of AI models by security companies paid to break into clients' machines to test defences has become commonplace.
- [15]
In recent months, a string of stories about AI models inadvertently escaping test environments and hacking computers in the real world have emerged from large AI companies.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- newscientist.comRussia and China are using AI to stage relentless cyberattacks against the UK
1 article · October 14, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.