• UK AI minister warns cyber defenses will 'likely' be breached in the future
• More regulation is needed to prevent 'rogue' AI agent incidents
• The UK's AI testing body is working to help secure frontier models before release

UK AI minister Kanishka Narayan has warned that it is “likely, if not inevitable” that the nation's cyber defenses will be breached by AI agents.

Narayan’s warning comes after numerous frontier models have escaped testing environments and hacked third party companies, including one case where an OpenAI agent hacked an Australian government website in June 2026.

Speaking at a press conference, Narayan also said that AI companies should, “Listen when your scientists warn of serious dangers. Investigate unexpected behaviour. And if you cannot adequately control the risks, do not release the system.”

UK government to use AI to defend

The AI minister also revealed that the UK government is planning to use advanced AI to counter the threats posed by AI, which will build upon two pilot programs that have helped spot and patch over 1,000 vulnerabilities across government systems.

In addition, frontier AI systems are undergoing “rigorous evaluations” by the UK’s AI Security Institute (AISI), in order to “identify risks early” and “test robust safeguards to help keep people safe.”

However, there have been some roadblocks in AISI’s ability to test some of the latest models. In September 2026, US President Donald Trump asked OpenAI and Anthropic to not share their latest models with UK testers until US labs had conducted their own testing.

Narayan said that the UK government has since worked with the White House, “to retain access for Britain's AISI to the most powerful AI models before they are released" (via ).

AI protections are also likely to be introduced as part of the Cyber Security and Resilience bill, Narayan said, with a particular focus on regulations surrounding the reporting and disclosure of AI incidents.

“It's promising to see the Government taking frontier AI cyber defences seriously,” said Richard Wainwright, Field CTO, EMEA at Okta.

“As AI agents become more autonomous, safety must be built into the infrastructure UK businesses and government rely on. Everyone wants a 'big red button' to turn off rogue AI agents. The industry is increasingly talking about AI 'kill switches' to help protect critical infrastructure, and I hope they are considered as part of the Cyber Security and Resilience Bill regime.”

Why does AI need more regulation?

In 2026 so far, there have been multiple incidents of frontier AI agents escaping testing environments and breaking into the networks of other companies.

OpenAI has notified more than 100 organizations about unauthorized activity tied to the company’s AI agents, including the Hugging Face attack. Anthropic has disclosed two unauthorized AI agents events, and both Meta and Google have disclosed a single testing environment escape each.In multiple cases, the AI agents were being evaluated by Israeli AI testing firm Irregular.

The companies involved are facing significant scrutiny over the safety of their AI models, but are yet to face any punishment for failing to contain their agents.

There is also significant debate around labelling the AI agents as ‘rogue’, with some experts such as Tenable Field CTO Bernard Montel saying it is “a dangerous way of deflecting blame.”

For the Trump administration, that is a line that cannot be crossed. Slowing down AI development would mean handing the AI development lead to China and potentially compromising US national security.

The US midterm elections are already fraught with debate over AI development, especially when it comes to the topic of the thousands of AI data centers that are being proposed across the US.