Compound Intellect
Home → Guides → Protecting assessment from AI

How to protect summative assessment from AI without banning it everywhere

By Basharat Ali, founder of Compound Intellect ·

Short answer: decide the boundary before the work is set, not after it is handed in. List the activities that are individually assessed, mark each one AI-free with the authority behind it (school policy, curriculum or a teacher's recommendation), schedule them in the term plan, and use AI on purpose in the practice that leads up to them. Detection tools after the fact cannot do this job.

Why detection is not a plan

AI-detection scores are probabilistic, and a wrong accusation lands on a student. Compound Intellect does not build AI detection or plagiarism scoring, by design. The reliable protection is structural: the assessed work is designed and scheduled so that it is done unaided, and everyone knows which work that is.

Four steps

  1. Find the assessed activities. Read the scheme of work for every activity that is individually assessed: write-ups, tests, coursework, presentations.
  2. Mark each one AI-free and name the authority. Most come from the school's policy ("no AI in summative assessment"). Some come from the curriculum, where the specification requires the student's own work. Label which.
  3. Put them in the calendar. A boundary teachers have to remember is a boundary that gets missed. In the term plan, the AI-free weeks should be visible, not in a footnote.
  4. Use AI deliberately before them. Practice is where AI can help: feedback on a draft the student wrote, critique of a design they made. Then the assessment checks what the student can do alone.

An example from a real course

In the Biology 10 course Compound Intellect is tested against, the Cell Structure and Transport unit ends with an individually assessed practical report. That report is AI-free, and the authority is the school's policy, which prohibits AI in summative work. Earlier in the same unit, students use AI to critique their investigation design after drafting their own controls. The unit is not "AI-free"; the assessment is.

What to tell students

Say which activities are AI-free and why, at the start of the unit. Students follow a rule they can see far more reliably than one they discover when a grade comes back.

Related: Where AI should not be used in a course · Student AI disclosure statement · AI in education statistics 2026

Questions schools ask

How can teachers stop students using AI in assessments?
Plan it before the work is set: identify individually assessed activities, mark them AI-free with the authority behind it, schedule them visibly, and design them to be done unaided, often in class.
Are AI detectors reliable enough to use in grading?
They give probabilities, not proof, and a false positive accuses a student. Compound Intellect does not build AI detection, by design; it plans the boundary before the work is set.
Does protecting assessment mean banning AI from the whole unit?
No. The assessed activity is AI-free; the practice leading up to it can use AI deliberately, with boundaries.
Try it on one course, for one term.

The pilot is one department, one term, free and with no card. You keep every plan, package and resource whatever you decide afterwards.

Talk about a pilotbasharat@compound-intellect.com