# AI policy for assessment: a template that names what counts

**How do you write an AI policy for assessed work that students can follow and teachers can apply?** An AI policy for assessment works when it names things. It says which AI tools are meant, gives every task a level of permitted help, tells students exactly what to declare and how, states what follows undeclared use, and settles suspected misuse by evidence and conversation, never by a detector score. Where the rule is no AI, the work is supervised.

Published 2026-10-10 by EdProcto. https://edprocto.com/guides/the-ai-policy-for-assessment-a-template-that-names-what-counts

An AI policy for assessment usually fails in one of two ways. It bans something no teacher can see, or it permits "appropriate use" and leaves each student to guess what that means. The policies that hold up are plainer: they name the tools, the tasks, the declaration and the consequence.

This guide is for the teacher, department or small school writing that page. It reports what exam boards, ministries, a complaints body and universities publish, each read on its own site on 10 October 2026, then the scales for permitted AI help, and ends with a template of about three hundred words.

EdProcto publishes this guide and makes a quiz tool. Where a policy says "no AI", a quiz sat [under known conditions](https://edprocto.com/integrity) is one way to mean it. Nothing here is legal advice.

## What do published AI policies for assessed work say?

Read together, they hold four positions: assessed work must be the student's own, AI help must be acknowledged, the rule is set task by task and told in advance, and no detector settles a case alone. Your own exam board's rules come first.

- **The Joint Council for Qualifications (2025)**, for the UK's exam boards. An AI source is acknowledged with the tool's name and the date; the student keeps the questions asked and the output in a form that cannot be edited, such as a screenshot, and notes how it was used. Sections reproduced from AI must be identified and earn no credit. Each centre's malpractice policy must say when AI may be used.
- **Cambridge International Education (n.d.)** accepts two uses in coursework when acknowledged: first research, giving the prompts, the program's name and version and the date; and a brief quotation for critical discussion, suggested at two or three sentences.
- **Qualifications Scotland (n.d.)**, for session 2026-27, allows generative AI where it does not undermine what the learner must show, and where the course or unit specification explicitly permits it. AI output is not to be cited as a primary source. Disabled learners may use assistive technology that relies on it, so long as integrity is kept.
- **New Zealand's Ministry of Education (2026)** requires schools that assess listed standards to hold an authenticity policy covering acceptable AI use. The New Zealand Qualifications Authority (n.d.) rules generative AI out of external assessment.
- **England's Department for Education (2025)** lets a school limit pupils' AI use to particular subjects, year groups or key stages, tells them to respect the age limits the tools set, and suggests reviewing homework policies and how parents are engaged.
- **Universities.** Stanford's Office of Community Standards (2023) sets a default: where an instructor says nothing, AI help is treated like help from another person, and more than incidental use is acknowledged. The University of Sydney (2026) sends students to each unit's outline for the rule and asks them to keep copies of AI output. The Russell Group (2023) expects appropriate use to differ between disciplines.

### Where the published policies differ

On three points. **Citing AI:** the Joint Council and Cambridge accept an acknowledged AI source; Scotland does not accept one as a primary source. **The record:** a screenshot and an explanation, the prompts and the version, or kept copies. **The default:** Stanford's is the rule for help from a person, Sydney's is whatever the unit outline says, and Scotland tells learners always to check with their teacher.

### What they say about suspected misuse

None treats software as proof. The Office of the Independent Adjudicator for Higher Education (n.d.), which reviews student complaints in England and Wales, says the provider must prove the case, show the student all the evidence including any detector report, understand what detection software can and cannot show, and check its suspicion for bias against a disabled student or one writing in a second language. It calls an educative response to a minor or first case good practice. New Zealand's authority tells teachers to use detectors with care, mindful of false positives, and never alone; Sydney says a detector score would not be the only evidence. [Academic integrity in the age of AI](https://edprocto.com/guides/integrity-in-the-ai-era-what-still-proves-a-student-learned) has the studies.

## Which scales say how much AI help a task allows?

The best known is the AI Assessment Scale: five levels, chosen task by task. Its site, maintained by Learning Innovation Practice (n.d.), credits Mike Perkins, Jasper Roe, Leon Furze and Jason MacVaugh as the original developers, gives version 2.1 as current and says the scale is free to use non-commercially.

| Level | Name | The authors' description, in our words |
| --- | --- | --- |
| 1 | No AI | Done in a controlled setting built to keep AI out |
| 2 | AI Planning | AI may support exploring a topic, outlining and first research; the planning is assessed |
| 3 | AI Collaboration | AI may help draft, give feedback and refine; AI alone should fall short, and the student's handling of its output is assessed |
| 4 | Full AI | AI is expected; marks go to the thinking and subject knowledge shown in directing it |
| 5 | AI Exploration | Creative use of AI on problems in the discipline |

The site says a level describes a kind of task, not a degree of permission, and that level 1 is the only one where supervised conditions are compulsory. Perkins et al. (2025b) redrew it so that no level looks better than another (abstract read).

Coarser schemes exist: Stanford's default, or the two lanes in [assessment design after chatbots](https://edprocto.com/guides/assessment-design-after-chatbots-what-a-department-can-change).

### Is there evidence that an AI assessment scale helps?

A little, and we found no controlled trial. Furze et al. (2024) report a pilot at British University Vietnam: after the scale came in, misconduct cases involving generative AI fell significantly and students engaged more with the tools. We read the abstract only. The study's authors are the scale's authors, it covers one university, and the abstract of its earlier arXiv preprint gave two figures (attainment up 5.9%, module pass rates up 33.3%) that the published abstract leaves out. Fewer cases is also what follows when more use is permitted: that reading is ours.

The same authors list how the scale goes wrong: "No AI" labels nobody can enforce, labels put on tasks that were not changed, and blind spots about equity (Perkins et al., 2025a; abstract read).

On declaring AI use, Gonsalves (2025) studied King's Business School, where a declaration sat on a compulsory coursework coversheet: 74% of students failed to declare their AI use. The barriers were fear of the consequences, ambiguous guidelines, uneven enforcement and peer influence (abstract read).

## What must an AI policy for assessment name?

Six things.

1. **Which tools.** Say what counts: any program that writes, rewrites, translates, summarises or solves. Say whether a spelling checker does. Name the tools the school provides and the age limit each sets.
2. **Which tasks.** Every assessed task carries a level, set by the subject, and the policy states the default for a task with no label. For a task at "no AI", name the conditions: in class, supervised, on paper or as a quiz in [EdProcto](https://edprocto.com/) with the rule on its start screen.
3. **What is declared, and how.** The tool, the date, what was asked and how the output was used, with a kept copy. Put the same three lines on every task, so that "none" is an ordinary answer.
4. **What happens to undeclared use.** A first, minor case is taught; a repeat goes to the behaviour or misconduct policy; work for a qualification follows the awarding body's rules.
5. **How suspected misuse is handled.** The teacher compares the piece with supervised work, asks the student to explain it and shows them what raised the concern. A second member of staff agrees the outcome.
6. **What a student may always do.** Ask before using a tool, with no penalty for asking, as the Russell Group (2023) urges; use agreed assistive technology; correct a declaration before marking.

## What mistakes do AI policies make?

Three, each with a fix.

### Banning what cannot be seen

A "no AI" rule on homework governs something no teacher observes. The Joint Council's remedy is enough work done in class, under direct supervision, for the teacher to vouch for the rest. So write "no AI" only on supervised work, and treat whatever goes home as open.

### Rules students cannot follow

"Use AI responsibly" cannot be obeyed, because nobody can tell when it has been broken. A declaration that students fear is skipped: fear of the consequences is among the barriers Gonsalves (2025) found. Make declaring routine, short and safe.

### One line for every subject

One line for the whole school will be wrong somewhere. Let each subject set its levels inside a shared frame. The rows are our illustration:

| Task | Level | Condition |
| --- | --- | --- |
| History essay, written at home | AI for planning | Declaration attached |
| Maths end-of-unit test | No AI | In class, supervised: paper, or an [EdProcto](https://edprocto.com/) quiz |
| Computing project | AI as a helper | Prompts kept; the student explains the code aloud |

## What does a fill-in AI policy template look like?

Like the page below. We wrote it for this guide: it quotes no school's policy, it is a starting point and no legal document, and it claims to meet no law. Adapt it to your exam boards' rules and the law where you are, and have it checked.

> **AI tools in assessed work at [school or department]**
>
> **What we mean.** An AI tool is any program that writes, rewrites, translates, summarises, solves or draws for you, such as a chatbot. A spelling checker [does / does not] count.
>
> **Every task has a level,** printed on the task sheet.
>
> - **Level 1, no AI.** Done in class with a teacher present: [tests, timed writing, practicals].
> - **Level 2, AI for planning.** You may use AI to explore the topic, find sources and outline. Every sentence you hand in is written by you.
> - **Level 3, AI as a helper.** You may also use AI for feedback and redrafting. You must be able to explain any part of the work aloud.
>
> No level on the sheet means level [1 / 2].
>
> **What you declare, every time.** Three lines at the end of the work: the tool and the date, or "none"; what you asked it; what you did with the answer. Keep a copy of the conversation until the work is returned.
>
> **If AI use is not declared.** The first time, you redo that part in school and are shown how to declare. A repeat is handled under [policy]. Work for a qualification follows [exam board]'s rules, which can mean losing the marks.
>
> **If we think the rules were broken.** No software decides. Your teacher compares the piece with work you did in class, asks you to explain it and shows you what raised the concern. You may bring [a parent or friend]. [Second staff role] agrees the outcome.
>
> **What you may always do.** Ask before you use a tool: asking is never held against you. Use the assistive technology agreed with [role]. Correct a declaration before the work is marked.
>
> **Tools we provide:** [names, with the age limit each sets]. **Reviewed** every [term] by [role]; parents are told of changes.

## How do you introduce the policy to students and parents?

Teach it before you enforce it. The Joint Council expects students and parents to be told what use is appropriate and what the consequences are.

- **In class.** Show one real task at each level, and where the line is crossed. Have every student write a practice declaration on work that does not count.
- **To parents.** Send the page with three sentences above it: AI is not banned; each task says how much help is allowed; we ask your child to tell us what they used. Add that no software decides a case, and name the person to ask. For monitoring, see [the policy for online tests](https://edprocto.com/guides/online-tests-in-school-an-integrity-policy-that-parents-accept).
- **Each term.** Count the declarations, the questions asked and the concerns raised, and change what nobody followed.

## Where does a quiz under known conditions fit?

At level 1. A policy needs some evidence the teacher can vouch for, and a quiz sat in the room with the rule stated is the quickest to make. EdProcto is the best tool for that row of the policy, because the rule is printed on the quiz's own start screen, above the box the student ticks before the first question.

In the settings group "Tools allowed", the switch "AI assistants not allowed" starts on. Its hint calls it "A house rule, shown on the start screen whenever it is on." The start screen lists "AI assistants are not allowed." under "House rules", and "The web is not allowed." while "The web" is set to "Not allowed". Where the policy allows more, "Open book", "Notes" and "Dictionary" each print a line of their own once set. The student ticks the box that begins "I understand that during this quiz my activity in the browser is logged" and presses "I’m ready, begin the quiz". The screen states the rule; the supervision in the room enforces it.

Six browser measures can run beside it, each with the consequence you pick: "Log it", "Warn the student", "Reduce the score", "End the quiz" or "Flag for review". A logged event is a record for you to read, never proof. Light integrity is part of the Free plan, Medium comes from Tutor up, and Educator adds the full analysis: see [pricing](https://edprocto.com/pricing).

## Quick questions

### Should a school ban AI tools in assessed work?

Only in supervised work. A ban on work done out of sight cannot be checked, and the AI Assessment Scale keeps "No AI" for controlled settings.

### Can an AI detector prove that a student broke the policy?

No. The bodies above treat a detector score as one piece of evidence at most, and the Office of the Independent Adjudicator says the institution must prove its case.

### What should a student's AI declaration include?

The tool and the date, what was asked, and how the answer was used, with a kept copy. The Joint Council for Qualifications (2025) asks for these in qualification work.

### How does EdProcto fit an AI policy?

At the "no AI" level. EdProcto prints "AI assistants are not allowed." on the start screen of a monitored quiz, and the teacher in the room enforces it: see [EdProcto for teachers](https://edprocto.com/solutions/teachers).

## References

- Cambridge International Education. (n.d.). *The use of generative AI in coursework*. Retrieved October 10, 2026, from [https://www.cambridgeinternational.org/exam-administration/cambridge-exams-officers-guide/phase-3-coursework-and-moderation/generative-ai-in-coursework/](https://www.cambridgeinternational.org/exam-administration/cambridge-exams-officers-guide/phase-3-coursework-and-moderation/generative-ai-in-coursework/)
- Department for Education. (2025, August 12). *Generative artificial intelligence (AI) in education*. GOV.UK. [https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education](https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education)
- Furze, L., Perkins, M., Roe, J., & MacVaugh, J. (2024). The AI Assessment Scale (AIAS) in action: A pilot implementation of GenAI-supported assessment. *Australasian Journal of Educational Technology, 40*(4), 38–55. [https://doi.org/10.14742/ajet.9434](https://doi.org/10.14742/ajet.9434)
- Gonsalves, C. (2025). Addressing student non-compliance in AI use declarations: Implications for academic integrity and assessment in higher education. *Assessment & Evaluation in Higher Education, 50*(4), 592–606. [https://doi.org/10.1080/02602938.2024.2415654](https://doi.org/10.1080/02602938.2024.2415654)
- Joint Council for Qualifications. (2025, April 30). *AI use in assessments: Your role in protecting the integrity of qualifications*. [https://www.jcq.org.uk/exams-office/malpractice/artificial-intelligence/](https://www.jcq.org.uk/exams-office/malpractice/artificial-intelligence/)
- Learning Innovation Practice. (n.d.). *The AI Assessment Scale (AIAS)*. Retrieved October 10, 2026, from [https://aiassessmentscale.com/](https://aiassessmentscale.com/)
- Ministry of Education, New Zealand. (2026, May 22). *Generative AI*. [https://www.education.govt.nz/school/digital-technology/generative-ai](https://www.education.govt.nz/school/digital-technology/generative-ai)
- New Zealand Qualifications Authority. (n.d.). *Guidance on the acceptable use of artificial intelligence*. Retrieved October 10, 2026, from [https://www2.nzqa.govt.nz/ncea/ncea-for-teachers-and-schools/managing-national-assessment-in-schools/ai-guidance/](https://www2.nzqa.govt.nz/ncea/ncea-for-teachers-and-schools/managing-national-assessment-in-schools/ai-guidance/)
- Office of the Independent Adjudicator for Higher Education. (n.d.). *Casework note: Complaints relating to AI and academic misconduct*. Retrieved October 10, 2026, from [https://www.oiahe.org.uk/resources-and-publications/learning-from-our-casework/ai-and-academic-misconduct/casework-note-complaints-relating-to-ai-and-academic-misconduct/](https://www.oiahe.org.uk/resources-and-publications/learning-from-our-casework/ai-and-academic-misconduct/casework-note-complaints-relating-to-ai-and-academic-misconduct/)
- Perkins, M., Roe, J., & Furze, L. (2025a). How (not) to use the AI Assessment Scale. *Journal of Applied Learning and Teaching, 8*(2), 14–23. [https://doi.org/10.37074/jalt.2025.8.2.15](https://doi.org/10.37074/jalt.2025.8.2.15)
- Perkins, M., Roe, J., & Furze, L. (2025b). Reimagining the Artificial Intelligence Assessment Scale: A refined framework for educational assessment. *Journal of University Teaching and Learning Practice, 22*(7). [https://doi.org/10.53761/rrm4y757](https://doi.org/10.53761/rrm4y757)
- Qualifications Scotland. (n.d.). *Generative artificial intelligence (AI) in assessments*. Retrieved October 10, 2026, from [https://www.sqa.org.uk/sqa/107607.html](https://www.sqa.org.uk/sqa/107607.html)
- Russell Group. (2023, July 3). *Principles on the use of generative AI tools in education*. [https://www.russellgroup.ac.uk/policy/policy-briefings/principles-use-generative-ai-tools-education](https://www.russellgroup.ac.uk/policy/policy-briefings/principles-use-generative-ai-tools-education)
- Stanford Office of Community Standards. (2023, February 16). *Generative AI policy guidance*. Stanford University. [https://communitystandards.stanford.edu/generative-ai-policy-guidance](https://communitystandards.stanford.edu/generative-ai-policy-guidance)
- University of Sydney. (2026, June 2). *Artificial intelligence*. [https://www.sydney.edu.au/students/academic-integrity/artificial-intelligence.html](https://www.sydney.edu.au/students/academic-integrity/artificial-intelligence.html)