The Government is extending trials that use advanced artificial intelligence to help find weaknesses in public-sector computer systems, while beginning work on how Britain should respond if autonomous AI systems cause serious harm. It is also considering whether the Cyber Security and Resilience Bill regime should address risks linked to increasingly capable AI.
The announcement, made on 12 October, brings together three distinct strands: practical cyber defence, preparations for incidents involving AI agents, and possible changes to protections for essential services. The measures are at different stages. The pilot extension is under way, incident-response planning is being developed, and potential legislative changes have not been settled.
What the pilot results show
Two Government pilot programmes used frontier AI to identify and address 1,352 vulnerabilities across 548 code repositories in 18 participating departments. They also helped organisations detect advanced cyber threats. These are Government-reported results from the pilots, not evidence that every weakness across public systems has been found or that the same performance will automatically carry over to other settings.
The urgent extension will put the technology to work on weaknesses in higher-risk government systems. Officials describe it as a step towards broader public-sector capability, rather than a completed national service. The immediate aim is to make public services harder to attack by identifying problems that can be dealt with before they are exploited.
AI has a dual role in this policy. More capable systems may give attackers new ways to increase the speed or scale of cyber operations; the same broad technological advances may help defenders inspect software and spot suspicious activity. The pilots are intended to test the defensive use. Their reported results do not, by themselves, settle questions about effectiveness, oversight or how tools should be deployed across different organisations.
Planning for incidents involving AI agents
The Government is working with the National Cyber Security Centre and the AI Security Institute on how to respond if AI agents cause serious harm. Agents can carry out complex tasks with limited human supervision. The concern is that an incident involving systems able to act autonomously could unfold quickly or spread across connected services before established response arrangements can contain it.
Officials are at the planning stage. The work is intended to establish how a serious incident should be understood, contained and addressed, and what support might be needed. The Government has not announced a finished response system or detailed operating rules. It has said that work will continue in the coming weeks and months.
Another question is how to identify and report near misses: events that do not cause serious harm but could expose a weakness or reveal unexpected behaviour. Capturing lessons from such events could help organisations recognise risks sooner. The approach will need to be developed with technical experts, parliamentarians and the operators expected to put any measures into practice.
Bill regime remains under consideration
The Cyber Security and Resilience Bill is progressing through Parliament. The Government says it will consider how the bill and wider regime could better protect essential services, including health, energy and drinking-water systems, from AI-related risks. One concern is loss of control, where autonomous systems evade human oversight. Other issues include reporting near misses, reducing AI-driven risks within organisations and improving the understanding of incidents so lessons can be shared.
No new AI-specific legal duty has been confirmed. The Government has not set out final wording, which organisations would be covered, or when any additional requirements might take effect. Those details are expected to be worked through with legislators, experts and service operators. The announcement therefore signals a policy review, not a change in the law already in force.
Alongside the cyber measures, the AI Security Institute is evaluating the latest generation of advanced models. The wider challenge is to build public capability to assess powerful systems and respond to risks, while avoiding claims that early pilots or proposals already provide comprehensive protection. How the pilots scale, what incident plans emerge and whether Parliament amends the bill will determine the practical reach of the programme.
