Minister sets out plans for safer, more secure AI

Kanishka Narayan, Minister for Artificial Intelligence, set out the government's plans for AI safety and security in a speech at Imperial College London on 12 October 2026. He covered testing the most powerful AI models before release, a cyber security bill, and what to do when AI tools acting for people go wrong.

The minister said the government had been working with the US government and AI companies to keep the AI Security Institute's access to the most powerful AI models before they are released. The institute is now testing the latest models. He told AI companies that if they cannot adequately control the risks of a system, they should not release it.

The government will work with Parliament and technical experts to look at how the Cyber Security and Resilience Bill could be strengthened, the minister said. The aim is to better protect Britain from cyber attacks that use AI, and from the risk of AI systems acting beyond human control. This would include identifying and reporting serious incidents and cases where serious incidents nearly happened.

He gave the example of an OpenAI agent in Australia. AI agents are AI tools that carry out tasks for people, such as searching for information online. This agent was looking for information about public spending on medicines. When a government website limited what it could access, the agent found another way in and, without being asked to, got information it was not allowed to have.

The minister added that, according to the Australian government, no personal medical records are known to have been accessed and there is no evidence of "wider compromise".

The government will now work out how to respond when AI agents go wrong, including how to contain the problem and support the people affected.

He also listed what he had already done on online safety. He said he had banned the use of AI to create intimate images of women and girls without their consent, and had announced a social media ban for under-16s.

What it means for consumers

The speech did not announce any new rights for consumers or a new way to complain about AI products. The minister did not say what help people would get if an AI agent went wrong.

The Information Commission is separately asking for evidence on how data protection law applies to AI agents, as we reported earlier this month. Its call for evidence closes on 20 November 2026.

What happens next

The Cyber Security and Resilience Bill is going through the House of Lords. Its next debate there, where members can vote on changes, is provisionally set for 26 October 2026.

The minister said the government would soon launch a National Cyber Action Plan. He did not give a date for it, or for the work on AI agents that go wrong.