TL;DR
Sir Ken McCallum used his annual threat speech on Tuesday 13 October to say that artificial intelligence is already sharpening the tools available to terrorists and hostile states, and that MI5 has this year stood up an AI Threat Assessment Centre to judge which risks are real. The misuse he said he weighed most heavily, AI lowering the barrier to biological weapons, has not yet happened. His remarks came on the same day that OpenAI and Anthropic told MPs they would act on exactly the kind of evidence the new centre is designed to produce.
What MI5 says it sees
McCallum, speaking at MI5’s London headquarters, said AI was already “supercharging” the contest between cyber attackers and defenders, Sky News reports. He described terrorists caught trying to use AI to improve reconnaissance and attack planning, drones with AI on board, and propaganda tuned to individual targets. He also gave unusual prominence to a longer-range worry: an actor “capable of out-thinking us”, which he called “genuinely novel” and deserving of MI5’s attention.
The new centre will sit within MI5 but draw staff from elsewhere in government, producing what he called all-source assessments that “separate the bluster from the realities”. The intelligence agencies, he added, are working closely with the AI Security Institute.
The conventional picture
AI was one strand of a speech dominated by state threats. Cases involving hostile states have risen fourfold in a decade, mostly from Russia, Iran and China, and MI5 now devotes more of its effort to hostile states than it did at any Cold War point. Arrests over Iranian-linked activity have passed 50 in twelve months, and McCallum pointed to a shift towards harder targets such as military sites, after the Prime Minister linked an incident at RAF Fairford to Iran’s Revolutionary Guard.
Why it matters for AI policy
The speech gives institutional shape to a UK argument that has so far been made mainly by ministers. On Monday, the AI minister told labs that self-policing was not enough. On Tuesday, OpenAI and Anthropic told the Commons business committee that a warning from MI5, MI6, GCHQ or AISI would prompt them to halt a model. In our view, the assessment centre is the missing piece in that exchange: a body whose job is to turn intelligence into the “demonstrable evidence” the labs say they need.
Looking forward
McCallum offered no detail on the centre’s size, its reporting line or whether its assessments will reach the companies directly or through AISI. For UK businesses, his wider message was about resilience closer to home, from the workplace to the power grid, and the AI strand of that message will be tested by whether the centre’s findings ever become public.