What happened.

A Brown University economics professor changed a take-home exam to an in-person format without AI. The take-home version averaged 96%, while the in-person version averaged 48.6%.

The change also coincided with 18 students dropping and nine students not attending the in-person exam. The supplied record does not explain why those students withdrew or were absent.

Why it matters.

The result suggests AI may be materially affecting performance on take-home assessments. Two larger studies from China and UC Berkeley are described in the research record as pointing in the same direction.

For schools, the question is not simply whether students use AI. It is whether take-home grades still show what a student can do independently under the conditions an exam is meant to measure.

What to watch next.

Watch for fuller results from the Brown experiment, including how the exams compared and whether participation changed for reasons connected to the format. Also watch for the underlying evidence from the China and UC Berkeley studies.

The record supports a concern about assessment design, but not a precise estimate of how much of any individual score was produced with AI. Simba Pool publishes this brief from an upstream research record and does not receive the upstream source-URL map.

What to watch

Watch for the underlying Brown, China, and UC Berkeley study materials, especially exam-comparison methods and participation details.

Receipts

Upstream references

Digest dated 2026-07-13 · upstream model claude-sonnet-4-6. Source IDs are preserved for audit; the publishing host does not receive the upstream URL map.

  1. 1
    969fdcb33dcde115e39e69299b889a217c38ca5cReference from the upstream research server

This quick brief was generated by Terra from a dated upstream research digest. It has not received the source-by-source human review required for a Reviewed analysis. Material limit: The supplied record is medium confidence, largely single-source, and does not include the underlying source URLs or enough methodological detail to establish causation.