The United Kingdom government has published a draft guidance document for civil servants outlining recommended practices for using artificial intelligence (AI) in government work. The document is titled Using AI ethically and sustainably and is currently a draft rather than a finalized policy.
Key contents of the draft
The under-600-word guidance collects several practical recommendations. It first stresses that users must follow any departmental AI policies, since those may mandate specific tools or procedures.
The text notes that how AI is used should depend on the nature of the task, so it is important to choose an appropriate model or tool. Citing examples reported by Tom's Hardware, the draft suggests that in some cases Gemini Flash could be preferable to Gemini Pro, and GPT Instant may be a better fit than GPT Thinking. The guidance distinguishes model types: models like GPT Thinking perform step-by-step reasoning before answering, while GPT Instant is optimised for everyday, fast tasks.
The draft also includes ethical guidance: results produced by AI should be verified before use. Verification covers not only factual accuracy but also ensuring generated text does not contain stereotypes or discriminatory language. Importantly, the document emphasises that the employee remains ultimately responsible for any content or decision that arises with the aid of AI, encouraging staff not to rely solely on automated outputs.
Sustainability and politeness
On sustainability, the guidance makes a notable recommendation: staff do not need to thank AI systems for their responses. The draft argues that such politeness is unnecessary because the models have no feelings and that extra interactions can impose a significant environmental burden. The document echoes prior comments from OpenAI: chief executive Sam Altman has previously pointed out that users' politeness can add tens of millions of dollars in operational costs to the company due to increased compute and energy use.
The guidance also advises using the fewest possible prompts and keeping instructions short and clear to minimise unnecessary computation.
Technical/access note
The draft references a prompt collection, but the linked resource is inaccessible when clicked, suggesting it may only be available on the government's internal network.
Overall, the draft aims to provide concise, practical advice to help government employees use AI "safely and sustainably," highlighting appropriate tool selection, verification of outputs, personal responsibility, and measures to reduce environmental impact.
