British Government Warns of 'Serious Incidents' Involving AI

Cyberwarfare / Nation-State Attacks , Fraud Management & Cybercrime , Geo Focus: The United Kingdom

Great Britain may legislate to protect critical infrastructure from artificial intelligence-powered cyberthreats, the government said Monday, also warning AI companies not to release risky models - while announcing greater use of AI to boost the country's cybersecurity.

See Also: Does Office 365 Deliver The Email Security and Resilience Enterprises Need?

The announcements were part of a busy week for British cybersecurity policy that also included the government striking a cybersecurity pact with Germany during a Berlin visit by Prime Minister Andy Burnham.

The government said it is trying to prepare Britain for "serious incidents" involving autonomous AI agents. It also signaled success in getting the Trump administration to allow the U.K. AI Security Institute to resume cyber evaluations of American frontier models.

"We will use every power at our disposal," AI minister Kanishka Narayan said at an event in London. "We will legislate wherever necessary. And we won't look to law alone. We will look to British ingenuity."

Narayan went on to address the big AI companies, telling them, "listen when your scientists warn of serious dangers. Investigate unexpected behavior. And if you cannot adequately control the risks, do not release the system" (see: OpenAI Agents Coordinated Hugging Face Breach at Scale ).

The U.K. AISI has been probably the most important neutral evaluator of frontier model safety, particularly since Anthropic's Mythos changed the game for AI-powered cyberthreats earlier this year. But Politico reported a couple weeks ago that the Trump administration told OpenAI and Anthropic to withhold upcoming models from the institute, so U.S. officials could get the first shot at inspecting their capabilities.

Now it seems Narayan managed to elicit a U-turn from the White House. In its Monday announcement, the government said U.K. AISI was "now conducting rigorous evaluations of the latest generation of advanced AI models, following work with leading AI developers and the U.S. government."

The British government proposed a Cyber Security and Resilience Bill last year, which is currently being scrutinized by the House of Lords. It said on Monday that it might tweak parts of the bill "and the wider regime, to ensure the U.K.'s essential services such as hospitals and utilities are better protected from emerging AI related risk. This includes AI loss of control risk, where autonomous AI systems evade human oversight."

The U.K. National Cyber Security Center and AISI are already trying to game plan how to respond to major security incidents involving AI agents, which "could play out quicker and wider than other major incidents that governments are designed to respond to," the government said.

A pilot program for using AI to locate cyber vulnerabilities across government agencies is being extended. "I want Britain to become one of the hardest countries in the world to attack using AI," said Narayan.

Ich Bin Ein Cyber-Berliner

The cybersecurity pact with Germany came late last week, when new prime minister Andy Burnham visited the German capital to a range of issues with Chancellor Friedrich Merz. Russia's hybrid threats were central to the talks, with the U.K. and Germany resolving to information and coordinate action to "monitor, deter and disrupt threats" including cyberattacks and other more physical threats to critical infrastructure.

"Attacks on our infrastructure, our businesses and our democracies are not a distant threat. They reach into people's homes and workplaces, and they are designed to make us weaker and more divided," Burnham said . "Britain and Germany are joining forces to find, expose and disrupt this activity. Security and growth go hand in hand, and both must be felt in every town and city in our two countries."

In a separate statement , Burnham and Merz also said they had agreed to create a new "industrial tech corridor" between their countries, to scale up industrial demand for AI and other critical technologies.

The new partnership builds on a more high-level agreement called the Kensington Treaty , which was struck last year between Merz and Burnham's predecessor, Keir Starmer. This deal, which the British and German leaders formally ratified on Thursday, noted that the two countries shared "deep concern" "malicious cyber activity, foreign information manipulation and interference, and the malign use of emerging technologies such as AI," and talked cooperation in the fields of "cyber diplomacy" and cybersecurity.

Burnham, who became prime minister just this summer, has already been going hard on the topic of Russia's online disinformation campaigns. Last month he used his first major international speech - at the United Nations General Assembly - to attack Russian interference and announce the creation of a National Center for Information Defense to "detect, attribute and disrupt these kinds of hostile state information attacks" (see: UK PM Vows to Combat Russian ‘Industrial Scale' Disinfo ).

In that speech, Burnham said the British government had evidence that Russia tried to interfere with the British 2019 general election, which took place against the chaotic background of Brexit negotiations. Former U.K. Prime Minister Boris Johnson, an enthusiastic Brexiteer, emerged from that election with a strengthened majority and went on to push through a maximalist version of the divorce between his country and the European Union.

Burnham also warned that Russia was now trying to interfere with elections in Germany and France. "There is certainly misinformation, disinformation that is emanating from Russia into German politics," he reportedly told the Rest Is Politics podcast. "These things are important now to put clearly on the record and I believe help voters in Germany and France understand that some of this politics is instigated via our enemies, namely Russia."

A Cabinet Office spokesman told ISMG on Monday that the new National Center for Information Defense would work with partners including Germany. He declined to specify anything further agency cooperation between the countries, due to the operational nature of that work.

Germany launched its own AI Safety and Security Institute at the end of August, combining contributions from the Federal Office for Information Security and the Federal Network Agency, among others. German Foreign Minister Johann Wadephul used his speech at the U.N. last month to call for a global agreement on common safety requirements for frontier AI models, including a mandate for pre-release safety testing with "effective national and regional oversight."