UK Plans New Defences Against AI Cyber Risks, What Was Announced

The government wants Britain to become "one of the hardest countries in the world to attack using artificial intelligence." That was the pitch from AI Minister Kanishka Narayan on 12 October 2026, as he set out a package of measures covering cyber defence, possible new legal protections and preparation for serious incidents involving autonomous AI systems.

Most of what was announced is a statement of direction rather than finished policy. Some of it is already under way. Some of it, by the government's own account, has yet to be worked out with parliamentarians, technical experts and the operators who would have to deliver it.

What the government announced on AI risks and cyber defence

The announcement came from the Cabinet Office, the AI Security Institute, the National Cyber Security Centre and the Department for Digital, Culture, Media and Sport. Narayan delivered it at Base One, Imperial College London's science and tech incubator, where he argued for a "whole society" approach to resilience.

His warning was blunt. As frontier AI systems grow more powerful and capable of acting autonomously, the consequences of failures or malicious use could become far more serious, he said. Britain, in his view, has to strengthen its defences now and prepare for serious incidents before they happen.

The government describes the package as a first step. Wider work is being led by the Prime Minister's AI Taskforce, which will spend the coming months focusing on the areas most affected by AI, from national security and economic growth to public services and the workforce. The Prime Minister told the UN General Assembly in September that the government would do whatever is necessary to protect British people from AI risks, and AI is due to sit at the heart of the UK's G20 Presidency next year.

Using frontier AI to find weaknesses before attackers do

The most concrete element is the extension of two existing pilots. According to the government, they have already helped identify and fix over a thousand potential vulnerabilities across government systems, and detect potential cyber threats. Ministers say extending the programme will protect everyday public services in the short term and support longer term adoption across the public sector.

A separate GOV.UK case study, published on 12 June 2026 by the Department for Science, Innovation and Technology and the NCSC, shows what this looks like in practice. The Government Cyber Coordination Centre, a partnership between the NCSC and DSIT, ran weekly in person hackathons in which teams used frontier AI to scan public code repositories across government, working with specialists from the AI Security Institute and the NCSC.

Teams were not told to use a single method. One ran each repository through a six stage AI agent pipeline, with each stage challenging the one before. Another used conventional scanning tools first and layered three model stages on top. A third department turned an audit spanning hundreds of services into five reusable, domain specific Claude Skills. In every case, the case study says, a human expert stayed in the loop on anything that mattered.

Participants identified 407 findings in total, including critical weaknesses that exposed services to authentication bypass, data exposure and remote code execution. Some were already understood and covered by compensating controls, while others were previously unknown. The case study says all critical weaknesses have been remediated and that no evidence of exploitation was found for any finding. The exercise ran across nine government organisations for a month and cost £13,000 in tokens.

The authors are candid about the limits. Agents produce candidate findings far faster than people can validate them, they write, and finding a flaw is not the same as fixing it, since every issue still has to pass through the patch pipeline. They also say the model matters less than the way it is used, with structure and triage doing most of the work. A second phase will add more departments and models, and extend the work from public code to closed source estates.

Could the Cyber Security and Resilience Bill cover AI incidents?

On legislation, the government is going no further than exploring. It says it will consider strengthening parts of the Cyber Security and Resilience Bill and the wider regime so that essential services such as hospitals and utilities are better protected against emerging AI related risks. That includes what it calls AI loss of control risk, where autonomous AI systems evade human oversight.

Officials will also look at how serious AI incidents and near misses can be identified and reported, and how lessons can be learned quickly to reduce the risk of future harm. No reporting threshold, deadline or penalty has been set out. The government says the detail will be worked through with parliamentarians, technical experts and the operators who would need to deliver any measures.

In his speech, Narayan said the government would "legislate wherever necessary" but would not rely on law alone, pointing instead to British scientific expertise, technological capabilities and institutions. His message to the companies building the most powerful systems was direct, listen when your scientists warn of serious dangers, investigate unexpected behaviour, and if you cannot adequately control the risks, do not release the system.

What happens if an autonomous AI agent causes a serious incident?

The announcement draws a distinction between AI agents and earlier tools. Agents can carry out complex tasks with little human supervision, and the government says a major incident involving them could play out faster and wider than the kinds of major incident its systems were designed to handle. Meeting that challenge, it says, will take new expertise and capabilities.

Work is under way with the NCSC and the AI Security Institute on how such incidents should be responded to. The government has not said when that work will conclude or what form the resulting response arrangements will take.

Alongside this sits the AI Security Institute's evaluation programme. Narayan confirmed that its scientists are now rigorously evaluating the latest generation of advanced AI models, following work with leading AI developers and the US Government. The June case study offers a hint of why that matters to government. It notes that recently released frontier systems, including Claude Mythos and GPT 5.5, brought a step change in cyber capabilities, and that AISI's evaluations show models improving quickly at cyber tasks.

The government also pointed to AI verification, the emerging field of checking that claims about how AI systems were developed, tested and assessed stand up to scrutiny. It says the UK has world leading expertise in the area and will set out more detail on support for it in the coming months.

What the AI cyber plan means for businesses, and what is still undecided

Cyber Minister Liz Lloyd said the aim was to put cyber security and AI preparedness at the heart of the UK's digital future, through scaled up AI pilots protecting public services and progress on the Cyber Bill. She said it was now business critical for organisations of all sizes to be on the front foot.

Earlier this year, ministers wrote an open letter to businesses urging action against AI enabled cyber threats. It asked them to take cyber security seriously at the top of the organisation, get the basics right through Cyber Essentials, and follow NCSC advice, including signing up to its Early Warning Service.

Several details remain open. The government has not published a timetable for changes to the Cyber Bill, a definition of a reportable AI incident, or a funding figure for the extended pilots. Those are matters the announcement leaves for later work.