A government wrote ten AI principles, then retired them in a year

Start here

Before an organisation lets staff loose on a chatbot, someone has to decide what ‘responsible use’ means in practice. In January 2024 the UK government published an answer for its own civil servants: ten principles, a list of promising uses, and a list of situations the guidance advises against. Anyone drafting workplace rules for the first time can read it as a template, not a mandate.

What the documents say

The Generative AI Framework for HMG, published 18 January 2024 by the Cabinet Office, Government Digital Service and Central Digital and Data Office, sets out ten principles running from understanding a model's capabilities and limitations, to keeping tools secure, to ensuring ‘meaningful human control at appropriate stages.’ It lists promising uses — drafting documents, summarising text, improving accessibility through translation — and separately advises against fully automated decisions on significant matters, high-risk uses affecting health or safety, and uses that require absolute accuracy or full explainability. The framework's own publication record states it ‘was withdrawn on 10 February 2025.’ Its replacement, the AI Playbook for the UK Government, describes itself as an expansion, retitled ‘to encompass a wider range of AI technologies’ beyond generative models alone.

Check this

Compare the framework's list of discouraged uses against a tool your own workplace is piloting: does the use involve a low-latency decision, a high-stakes outcome, or a need for a fully explainable answer? The framework's own text names these as the situations where generative AI is least suited, regardless of how well a demonstration performs. That is a checkable boundary, not a matter of opinion about the tool's quality.

What holds and what fails

The principle that holds up independent of any one model's improvement is the separation between promising and discouraged use categories — a distinction about the task, not the tool. What failed, on the document's own timeline, was the framework's shelf life: thirteen months from publication to withdrawal, replaced by broader guidance as the technology moved past ‘generative AI’ as a category. Editorially, that turnover is itself a lesson: a workplace policy this specific to one year's tools should expect the same fate, and ought to say who is responsible for revising it.

• Read the ten principles against your own team's current AI use, one by one.
• Check whether your workplace's policy has a review date, the way this framework eventually needed one.
• Note which of the framework's discouraged categories, if any, describes a task you are automating.

A principles document is a snapshot of an institution's judgement at one date; this one was current for about a year. Reading it in September 2026 means reading a retired policy for the reasoning it recorded, not for instructions still in force.