IT Home On October 12, according to the Guardian, Britain will strengthen its cyber defense and calls on top artificial intelligence research companies to take more proactive measures in system security governance.
UK AI Minister Kanishka Narayan is expected to officially announce the above policy proposals in a speech in West London.
Narayan will call on leading artificial intelligence research and development companies to take more concrete measures to ensure system security. According to IT Home, Narayan said that the UK must not completely rely on technology giants to “self-regulate” its models, and emphasized that more self-governing and controllable AI infrastructure must be built in the UK.
Further reading
According to the UK government’s official website, Karishka Naran was appointed Minister for Artificial Intelligence on July 20, 2026. He will also serve in the UK Cabinet Office and at the Department of Business, Innovation, Science and Trade (BIST). Previously, from September 7, 2025, to July 20, 2026, he served as Deputy Minister for Science, Innovation and Technology (DSIT) in the Parliament. In July 2024, he was elected as a member of the Welsh Glamorgan constituency.
Among the responsibilities of Naran’s Cabinet Office is the Artificial Intelligence Security Institute (AISI). This institution aims to establish a evidence base for artificial intelligence risks, providing a reference for government decisions. According to information disclosed by the UK Parliament in April 2026, AISI has collaborated with leading AI companies such as OpenAI, Google DeepMind, and Anthropic to test over 30 models. The findings from these tests will lead to actual changes before the models are deployed.
On December 18, 2025, AISI released its first “Frontier Artificial Intelligence Trends Report”. The report shows that the AI system capabilities have improved rapidly: in the field of network security, the success rate of models completing apprentice-level tasks has risen from less than 9% in 2023 to about 50% in 2025. In 2025, for the first time, a model was able to complete expert-level network tasks that required up to 10 years of experience. In the field of software engineering, the success rate of models completing programming tasks that take about 1 hour has increased from less than 5% two years ago to over 40%. For some network tasks, the duration required for AI to complete them without human guidance doubles approximately every 8 months. The report also noted that the protection measures used to ensure AI operates as expected are improving. The time it takes for red teams to find a “general jailbreak” method has been extended from minutes to hours, approximately 40 times longer.
In strengthening the domestic AI infrastructure, the UK government has previously taken a series of measures: establishing a Sovereign AI venture capital fund worth 500 million pounds, with the first investment going to the AI infrastructure company Callosum; releasing the AI Hardware Plan to support UK companies developing AI chips and semiconductor technologies; issuing the Cyber Growth Action Plan to boost the country’s cybersecurity industry, and expanding the scale of the National Security Strategy Investment Fund. The British government believes that the majority of AI systems should be regulated during use.