Skip to main content
Complimentary sampleNo account required← Entrance
The Library Brief · sample17 Jul 2026 · Complimentary edition · 4 items · 6 min read

Assessment after AI: the question schools can no longer defer

The tools that promised to police AI use are proving unreliable. That is uncomfortable—but it reopens a better question: what kind of assessment is actually worth setting?

AGCurated by Alex Gray

For two years, the reassurance offered to worried teachers was simple: if pupils use AI to write their essays, a tool will catch it. That reassurance is falling apart. Detectors are probabilistic, and even a modest error rate becomes unacceptable when the result could be an accusation of misconduct.

The change is not simply technical. More leaders are deciding that a detector score cannot carry a disciplinary decision on its own. Once that is accepted, the ground shifts: the real question is no longer how to catch invisible tool use, but what kind of assessment is worth setting when a machine can complete the task unseen.

That is not a loss of control. Detection was doing the policing—imperfectly—while allowing assessment design to escape scrutiny. Removing that false certainty creates room for better evidence of learning: annotated drafts, short spoken defences, supervised checkpoints and honest conversations about how a pupil arrived at an answer.

The thread through this Brief is what can replace detection without creating a marking mountain. Begin with the evidence, pilot the ten-minute protocol, then use the policy checklist to make the change defensible.

The rest · curated items

4 items

The Library

Where the evidence on AI-writing detection actually stands

Independent evaluations find substantial reliability problems, while research has also identified a particular risk of misclassifying writing by non-native English speakers.

Why it matters for schools: It gives a school a defensible basis for treating detector output as a prompt for review—not proof of misconduct.

Read the detector-bias study

The Lab

A spoken-defence protocol you can run in ten minutes

Ask a pupil to explain one decision, challenge one claim and connect the work to something taught in class. Record a simple judgement: secure, partial or not yet evidenced.

Why it matters for schools: It turns a vague call for “more oral assessment” into a small pilot a department can run next week.

Use the three-question protocol

The Library

The clauses an AI-use policy needs once detection is off the table

State what evidence may trigger a review, what other evidence must be considered, how pupils can explain their process and how they can contest an automated judgement.

Why it matters for schools: Clear review and appeal language matters more than a list of tools when a decision is questioned by a pupil or family.

See Jisc’s contestability principle

The Journal

What we lose when we stop policing prose

The useful question is whether a pupil is thinking with AI or using it to escape the work of thinking. That distinction belongs in pedagogy, not in a detector score.

Why it matters for schools: It offers constructive language for the staff conversation that follows when old certainties are removed.

Read the full DEEP essay

From The Lab · use it tomorrow

The ten-minute spoken defence

  1. 01Which decision in this work was most important, and why?
  2. 02Which claim would you defend if I challenged it?
  3. 03Show me where this connects to something we studied in class.

Record only whether understanding is secure, partial or not yet evidenced. The conversation is corroborating evidence—not a second examination.

Sources

This is what members receive three times a week.

Each Brief filters the noise, checks the evidence and explains what the development means inside a school.

Explore the four rooms →