The UK Department for Science, Innovation and Technology has issued new guidelines for the procurement and use of artificial intelligence within the public sector. The policy mandates that officials must first evaluate the necessity of using AI tools before deployment, emphasizing that traditional software or spreadsheets may suffice for many administrative tasks.
When AI is deemed necessary, the government instructs agencies to prioritize the smallest and least complex models capable of completing the task. This approach aims to avoid the unnecessary use of advanced, large-scale language models, which often involve higher computational costs and energy consumption.
The guidelines also highlight concerns regarding data privacy and technical sovereignty. By favoring lightweight or open-source models that can be deployed locally, the government seeks to minimize reliance on third-party cloud services and reduce the risk of sensitive public data being exposed to external platforms.
This strategy is designed to optimize public spending and align with environmental goals by reducing carbon emissions associated with large-scale model training and inference. The policy serves as a framework for balancing technological adoption with fiscal responsibility and security requirements.
