Summary

  • The UK announced £60m for its National Cyber Shield and £70m for a new National Exposure and Real-Time Vulnerability Centre on 24 October.
  • A £25m challenge fund sits inside the Shield allocation. The government will also explore a regulatory sandbox for agentic cyber security.

Spotting a hole in a network is one job. Closing it while hospitals, utilities or payment systems are using the network is another. Britain's new cyber spending announcement puts £130m behind that distance between seeing and doing. The invoice is easier to write than the permission slip.

The Treasury announcement assigns £60m to a National Cyber Shield that would use agentic AI in defence. Its £25m challenge fund is part of that allocation. A separate £70m National Exposure and Real-Time Vulnerability, or NERV, Centre is meant to use AI to find vulnerabilities across critical infrastructure and the wider economy. Ministers will scope a sandbox for deploying agentic security in critical infrastructure, government and industry. The announcement gives no launch date or operational charter for NERV.

The NCSC's July blueprint supplies the sharper question. It envisages agents that scan critical UK IP ranges, share warnings across organisational borders and eventually automate measures such as blocking malicious domains. It also says an organisation's agents run under that organisation's control and authority. National reach, in other words, still has to pass through local keys.

NCSC's September analysis describes why. An automated scan may observe; a firewall rule can interrupt a service; code execution can reach further. The agency weighs potency, scope, criticality, confidence in rollout and recoverability. A clever detector does not acquire the right to break a hospital appointment system merely because it found a suspicious packet.

The money is real, and so is the unresolved handoff. Cyber Shield's public blueprint aims to move from discovery to remediation, first with network defenders across government and critical sectors. The spending plan now needs an answer that operators can actually sign: which action can an agent take, on whose instruction, against which system, and who can reverse it?

Sources