UK renames its AI Safety Institute as the AI Security Institute

Background

The AI Safety Institute was an existing UK body. On 14 February 2025, the Secretary of State for Science, Innovation and Technology announced at the Munich Security Conference that it would be renamed the AI Security Institute.^[1] The Department for Science, Innovation and Technology's annual report also records that the institute was relaunched and renamed in February 2025.^[2] The change was therefore both a new name and a restatement of the institute's public mandate.

The new name and focus

The secretary said the new name better reflects the institute's focus on serious AI risks with security implications. He gave specific examples: how the technology can be used to develop chemical and biological weapons, carry out cyber-attacks, and enable crimes such as fraud and child sexual abuse.^[1] These examples place the institute's work in the area of severe harm and misuse, rather than in the full range of social and ethical questions raised by AI. In public policy, safety and security are often used to mark different emphases: safety can cover a broad set of harms, while security points more directly to risks that can be exploited to cause harm. The announcement drew its examples from that security side of the distinction.

Scope and boundaries

The secretary also set a boundary around the institute's work. He said it would not focus on issues such as bias or freedom of speech, which his department would address through other initiatives.^[1] That statement is important because it separates the institute's security-focused technical work from wider policy questions about how AI affects society. It also shows that the rename was not meant to make the institute the home for every AI-related concern. Instead, the institute is directed toward risks that have a security dimension, while other parts of the department handle other issues.

Testing AI models and the criminal misuse team

The secretary instructed the institute to take a leading role in testing AI models wherever they come from, open or closed.^[1] This wording matters because it says the institute's testing role should not depend on where a model was developed or whether it is open or closed. The institute is therefore meant to examine models across those categories, rather than only a narrow set of systems. He also launched a new criminal misuse team within the institute that partners directly with the Home Office.^[1] That team adds a direct link between the institute's technical work and the Home Office. The combination is notable: the institute is given both a testing role and an operational route for addressing criminal misuse.

What the change means

The rename shifts the institute's public emphasis from safety to security. The annual report confirms the organizational change, recording that the AI Safety Institute was relaunched and renamed as the AI Security Institute in February 2025.^[2] The announcement set out the direction: serious AI risks with security implications, model testing across open and closed systems, and a new team focused on criminal misuse with a direct Home Office partnership.^[1] Taken together, these points describe a mandate rather than a set of results. They show where the institute is expected to concentrate its effort and how it is meant to connect technical assessment with security and crime policy.