Manchester now labels every assessment with one of four AI categories

What changed on 1 September

The AI in Teaching and Learning Policy is a seven-page document that applies to undergraduate and taught postgraduate students and to the staff who teach them. Its document control box gives the approving body as Senate, the approval date as April 2026, the implementation date as September 2026 and the next review as September 2027. It records that the policy supersedes nothing, which makes it the first of its kind at the university.

The policy describes its purpose as supporting "consistent, transparent assessment expectations" while enabling appropriate use of AI. A staff news item published on 1 June 2026 said it was developed by academic colleagues, professional services colleagues and the Students' Union, and described it as "a workable starting point - not a final answer".

The four categories

The definitions below are taken from section 2.1.1 of the policy. The category decides what information an assessment brief has to contain.

The line between B and C is the one most likely to cause confusion, and the university's own staff say so. In an update to the Faculty of Humanities dated 25 August 2026, Mark Carrigan, the faculty's associate dean for digital education, wrote that there is "clearly a degree of ambiguity" between the two. His suggested test is whether AI use would change what the student is saying or only how they are saying it. Category B covers the second, which he compares to the help a proofreader might offer.

Category D is expected to be rare. The June staff notice said AI-Integrated assessments would apply to "a very small minority", and that optional units using them had to be flagged before course selection opened in August so students could make an informed choice. Carrigan's update said the faculty expected only a handful, because D is the one category where students cannot opt out.

Other rules in the policy

• Declaring AI use: section 3.0.2 says that within assessed work, a student's failure to document use of generative AI "could be a form of plagiarism" contrary to the Academic Malpractice Procedure. Failure to follow an assessment brief is handled case by case and may be dealt with through marking.
• Uploading course materials: students must not upload question papers, handouts or other copyrighted resources to external AI tools.
• Lecture recordings: where a recording of a taught session is available, students may upload it to their university-provided Microsoft 365 Copilot account.
• Group work: the unit lead must state the category and explain how the group should approach and record its use of AI. Students need agreement from all group members before recording or transcribing group conversations.
• Staff: only the university-provided Microsoft 365 Copilot service may be used for student-identifiable or confidential data.
• Wellbeing: the policy says Copilot is not a suitable substitute for therapy, medical advice or academic advice.

How it compares with other models

Manchester's framework is a variation on the traffic-light systems that, in Carrigan's words, have become commonplace: prohibition (red), context-specific use (amber) and encouraged use (green). It adds a fourth category. In a post on the university's teaching and learning blog in June, Carrigan wrote that the problem with traffic lights is that most assignments end up categorised as amber, so the question of clarity is pushed elsewhere. He expects most Manchester assessments to fall into the two middle categories, and accepts that the extra category brings a risk of ambiguity of its own.

Other institutions have gone the opposite way. The University of Bath's guidance for this academic year replaces its A, B and C categories with two lanes from September 2026: a closed lane where generative AI is not permitted and the assessment is designed to prevent it, and an open lane where its use is authorised. We covered that change in our report on Bath's two-lane model.

The difference matters because Categories A and B depend on students following a rule in work that may be done unsupervised. Carrigan does not present the policy as a fix. He calls AI in assessment a "wicked problem" with "only better or worse options".

What this means for students and teachers

• Manchester students: look for the category on each assessment's Canvas page and in the brief. If it is missing or unclear, ask the unit lead in writing before you start.
• Manchester students: in Categories B, C and D, record what you used AI for. The policy treats undocumented generative AI use as potential plagiarism.
• Manchester students: do not paste exam papers or handouts into outside AI tools. The policy prohibits it on intellectual property grounds.
• Students elsewhere: the same word can mean different things at different universities. 'Permitted' at Manchester still comes with a list of acceptable uses.
• Teachers: the B and C boundary needs examples. A test such as 'does it change what is said, or only how' is easier for students to apply than a category name.

FAQ

What happens next

The university has said it will review how the policy is working after the first semester, and the policy requires faculty teaching and learning committees to look at its impact at least once a semester. The first real test will be whether students find the B and C labels clear when coursework deadlines arrive. We will update this article when the university publishes the outcome of its review.