Exam rules default to no AI unless a student is told otherwise

Start here

A student wondering whether an assistant is allowed for a specific piece of work is asking the question the wrong way round. Two UK institutions that set exam and assessment rules — the qualifications body for GCSEs and A-levels, and a research university — both frame the real question the opposite way: is AI use specifically permitted for this assessment, in writing, in advance?

What the documents say

JCQ's guidance on AI use in assessments, retrieved 16 September 2026, states that under its General Regulations for Approved Centres, ‘teachers and assessors must only accept work for qualification assessments which is the student's own,’ and that a student who submits AI-generated material as their own ‘will have committed malpractice’ and ‘could attract severe sanctions.’ Any reproduced AI-generated sections must be identified by the student, and identified sections cannot count toward demonstrating the marking criteria independently. Oxford's guidance on safe and responsible use of GenAI, last updated 27 April 2026, states the rule from the opposite direction: ‘any use of AI in work submitted for assessment is unauthorised unless you are specifically told differently in writing, in advance of the assessment,’ and that unauthorised use ‘constitutes cheating and plagiarism under University rules,’ with penalties including failing the assessment or, in serious cases, expulsion. Oxford adds that permission granted for one assessment does not carry over to the next.

Check this

Before using any assistant on graded work, find the specific written instruction for that assessment — not a general school or department policy, and not what was permitted last term. Both documents put the burden on the student to locate that permission, and both treat its absence as a no by default rather than a grey area to interpret generously.

What holds and what fails

What holds across both is the default-to-no structure and the requirement that any permitted AI contribution be identified rather than blended invisibly into a student's own prose. What differs is enforcement detail: JCQ describes a national exams-malpractice process shared across awarding bodies, while Oxford's is a single university's conduct process with its own penalties. Editorially, the practical lesson generalises past these two bodies: an assessment's silence on AI is not permission, in the pattern both documents establish, and a student should assume the stricter reading until told otherwise in writing.

• Locate the written AI policy for your specific assessment, not a general one, before you start.
• If any AI-generated material appears in your submission, identify it explicitly rather than blending it in.
• Ask what happens to marks for identified AI-generated sections before assuming they still count.

Both documents are guidance current as retrieved, not case law, and both apply within one country's or one institution's system. What they share is worth carrying anywhere: the presumption runs against undisclosed AI use, not in favour of it.