Britain's government urged developers to conduct safety verification, saying artificial intelligence (AI) models should not be launched if risks cannot be sufficiently controlled.

Cryptopolitan, a blockchain media outlet, reported on Sunday that UK AI Minister Kanishka Narayan (카니시카 나라얀) warned frontier AI labs not to release models if they cannot adequately control risks.

Speaking at an event at Imperial College London, he said, "We must listen when scientists warn," and added, "If you cannot control unexpected risks, do not launch the system itself." He said safety verification should take priority over release even as competition to develop AI intensifies.

Britain's government is also continuing pre-checks of models. Narayan said the UK AI Security Institute is evaluating the latest generation of models, including Google's Gemini 4 Argon. He described this as a process to identify risks early and confirm that safeguards actually work.

He also stressed that the government is maintaining advance access needed for model evaluations. Narayan said he has worked in recent weeks with the U.S. government and AI labs for this. Reports previously said the U.S. government asked OpenAI and Anthropic to delay providing new models for UK testing, but the UK government has not officially confirmed that access for its domestic institutes was lost.

Britain's government is also using AI in cyber security responses. Narayan said the government used frontier AI in 2 pilot projects to find and remedy more than 1,000 vulnerabilities in public services. It also shortened the patch period for already known vulnerabilities to 8 days from 50 days through a central monitoring service. With AI being used for both cyber attacks and defence, it is expanding its response system across government.

It is also pursuing regulatory work in parallel. The UK government plans to push a Cyber Security and Resilience Bill with parliament and to establish a reporting system for serious AI incidents and incidents that nearly led to real harm. It is also reviewing measures for incident management and damage support for agentic AI that acts autonomously.

Narayan said Britain will treat AI as a core agenda item if it takes the presidency of the Group of 20 (G20) next year. The government also plans to release its first draft technical standards on Monday afternoon, with the AI Security Institute taking a central role in related work. The plan is to broaden the scope of AI safety management, from pre-evaluation of the latest models to strengthening public-sector security and establishing an incident reporting system.