Three AI policy templates disagree on one basic structure

The classroom note

A head teacher told to "write an AI policy" this term usually starts by searching for someone else's. Reading three of the templates and department documents actually available, as retrieved on 16 September 2026, shows they agree on some structural choices and disagree on others, which matters before any borrowed language gets pasted into a governors' meeting.

What the evidence says

The TeachAI toolkit, published in 2023 by a coalition including Code.org, ETS, ISTE+ASCD and Khan Academy, states plainly that its examples "are not meant to be copied verbatim," offered only to anchor discussion, and organises its recommendations around what it calls "the seven Principles for AI Guidance," alongside separate sections for reviewing existing policies and engaging parents. The Australian Framework for Generative AI in Schools, published in 2023 by a national Taskforce, instead sets out six named principles - Teaching and Learning, Human and Social Wellbeing, Transparency, Fairness, Accountability, and Privacy, Security and Safety - as a single formal document adopted across jurisdictions. The UK's Department for Education guidance, published 29 March 2023 and updated 12 August 2025, takes a third structural approach again: rather than a standalone framework, it recommends schools "review homework policies, and other types of unsupervised study" and fold AI into policies that already exist.

The implementation question

The disagreement that matters most for a governing body is not the wording of any principle but this structural choice: one new standalone policy, or amendments threaded through academic integrity, safeguarding, data protection and homework policies that already exist and are already approved. TeachAI and the DfE guidance lean toward the second path, treating a fresh AI-specific document as optional; Australia's framework took the first path nationally, in part because it needed one text multiple state systems could adopt together. A school combining these templates has to pick one shape before borrowing language, since a Teaching-and-Learning principle from one document and a homework-policy amendment from another do not automatically fit inside a single coherent document without editing.

What holds and what fails

What holds: all three documents agree that guidance should name specific, checkable practices - citation, risk assessment, existing-policy review - rather than a general statement of enthusiasm or caution. What fails: assuming any one template is complete on its own; each was built for a different governance structure, and none claims legal force in a jurisdiction it was not written for. This is an editorial reading built from three primary documents, not a template itself, and it is not legal advice about what a specific governing body must adopt.

• Are we writing one new standalone AI policy, or amending existing policies, as the DfE guidance suggests considering?
• Which named principles or practices are we borrowing, and have we adapted the wording to our own context rather than copying it?
• Does our policy include a plan to engage parents and review the document, as TeachAI's toolkit and the Australian framework both recommend?

None of these documents is a template to copy. Each is evidence of a different, defensible way to structure the same underlying decision.