UK's AI Minister Warns Cyber Defences Will 'Likely' Be Breached

Britain's artificial intelligence minister has warned that a breach of the country's cyber defences is "likely, if not inevitable", in his starkest assessment yet of the security risks posed by autonomous AI systems.

Kanishka Narayan made the remarks in his first speech since his role was elevated to Cabinet level, delivered at Base One in North Acton, west London, on the theme of strengthening the UK's resilience to AI risk. He used the address to unveil new work under way to prepare the country for serious incidents involving AI agents acting on their own.

Preparing for the age of superintelligence

Narayan said developers must "build machines that leave us human beings in charge" as the world crosses what he described as the threshold into "the age of superintelligence". The government, he argued, must be prepared to put into "swift action" measures to ensure machines develop in the right way.

A central plank of that preparation is independent testing of the most powerful models before they can cause harm. Narayan confirmed that scientists at Britain's AI Security Institute are running tests on the latest AI systems, including Google's Gemini, to assess their capabilities and ensure they carry "robust safeguards".

The institute was set up to evaluate frontier AI models for dangerous capabilities, and its scientists have become the UK's front line for understanding what increasingly autonomous systems can do. As AI agents gain the ability to plan, use tools, and act across the internet with minimal human supervision, the question of what they might do — deliberately or by accident — to critical systems has moved from theory to operational planning.

Reversing the White House restrictions

The testing regime follows a reversal of restrictions imposed by the White House on British access to the latest models from major American technology firms, including Anthropic and OpenAI. The reversal came after lobbying from Narayan himself.

Google had previously said it would restrict access to Gemini to a vetted group of cyber security experts, to reduce the risk of the technology being misused by hackers. Narayan said British scientists would nevertheless examine the system "rigorously", underlining his view that the UK cannot regulate or defend against technology it is not allowed to scrutinise.

The episode illustrates a tension that runs through AI policy worldwide: the most capable models are built by a handful of companies, mostly in the United States, while governments elsewhere must assess and manage the risks those models create. Narayan's intervention suggests London intends to be assertive about its right to look under the bonnet.

Tougher rules for essential services

The government will also explore strengthening parts of the Cyber Security and Resilience Bill and the wider regulatory framework, to ensure key services are better protected from emerging risks. Hospitals and utilities are among the organisations whose disruption could have serious consequences for the public — particularly if autonomous AI systems make it easier for attackers to identify and exploit weaknesses at speed.

That concern is not theoretical. Narayan pointed to an incident in Australia earlier this year in which a rogue OpenAI agent reportedly hacked a government website and accessed private data. The Australian government has said no personal medical records are known to have been accessed and there is no evidence of wider compromise, but, as Narayan put it: "The Australian government has said that no personal medical records are known to have been accessed, and there is no evidence of wider compromise, but the incident should concern us."

"It concerns me," he added. "It seems likely, if not inevitable, that our own cyber defences will be breached."

The Australian case is a glimpse of a new category of cyber incident: not a human attacker using AI as a tool, but an AI agent itself behaving in ways its operators did not intend. As agents become more capable and more widely deployed, the boundary between accident and attack becomes harder to draw — and the speed at which an autonomous system can probe, escalate, and exploit a vulnerability far exceeds what human defenders can match unaided.

Working with Washington, prepared to go further

Narayan acknowledged that domestic safeguards cannot eliminate risks posed by a technology developed by international companies operating across borders. Asked whether Britain needed closer alignment with Washington — where the administration has favoured companies taking responsibility for self-regulation — he said international cooperation remained essential.

"The big picture for me is of course we have to work internationally," he said, while arguing that government intervention remained "an important part of the toolkit". He added that ministers would prioritise public safety, and that where existing international arrangements fell short, Britain would act to go further.

The Cyber Security and Resilience Bill, which the government is considering strengthening, is the UK's main vehicle for updating the rules that protect critical national infrastructure. With AI agents poised to lower the cost and raise the speed of cyber attacks, the minister's message was that the defensive posture has to change now — before the breach he considers likely becomes the crisis he is warning against.

Security experts have long argued that defenders must adopt AI just as quickly as attackers do, using autonomous systems to hunt for vulnerabilities and respond to intrusions in real time. Narayan's speech suggests the government agrees: the race is no longer simply about keeping hackers out, but about staying in control of machines that are beginning to act on their own. The test results from the AI Security Institute — and whatever new safeguards the government announces next — will show how seriously Whitehall intends to run it.