What counts as academic misconduct when using AI?
Broadly, presenting AI-generated work as your own, and failing to declare AI use where the institution requires declaration — but the specific rules vary by institution and by assignment, and that variation is itself the main source of trouble.
Why this is treated seriously. Assessment certifies that you can do something. Submitting work you did not produce defeats that, regardless of the tool used — which is why most institutions treat it under existing plagiarism and contract cheating rules rather than creating a wholly new category.
What is generally prohibited:
Submitting AI-generated text as your own writing.
Using AI where a specific assessment forbids it, which is common for language, coding and writing assessments where the skill being tested is the thing AI would perform.
Failing to acknowledge use where acknowledgement is required.
Fabricated citations, which is a distinct and serious offence — and AI tools generate plausible but non-existent references frequently enough that this has become a recognisable marker.
What is generally permitted, though you must check:
Explaining a concept you then write about yourself.
Grammar and spelling assistance, which long predates AI.
Generating practice questions, structuring revision, and summarising your own notes.
Code assistance where the assessment permits it, which many computing courses now explicitly do.
The frequent tiered approach: many institutions now classify each assessment — AI prohibited, AI permitted with acknowledgement, or AI integral to the task. Read the brief for each assessment, since the answer genuinely differs between modules.
On detection. AI detection tools have high false-positive rates and have been shown to disadvantage non-native English writers particularly. Institutions have therefore moved away from relying on them alone, and cases are typically built on process evidence — version history, drafts, viva questioning about the work, and inconsistency with other submissions.
If accused, the ability to show your process is the strongest defence — which is a reason to retain drafts, notes and version history routinely.
The safest approach: declare use, keep your process, and assume the rule is per-assessment.