Detection is the wrong model.

Detection treats every student like a suspect. It scans for what looks like cheating after the work is done, and the published evidence says it does not do that reliably. A 2023 test of 14 detection tools, Turnitin among them, concluded that they were “neither accurate nor reliable” and that content obfuscation techniques significantly worsened their performance. A study in Patterns found that GPT detectors frequently misclassify writing by non-native English speakers as AI-generated.

The model is backwards. You can’t police your way to academic integrity. You build it — by teaching the process, making the work transparent, and giving students tools that show their thinking instead of hiding it.

That’s the bet behind Wonders: AI summaries link back to real sources, the process stays visible to supervisors, and students come out sharper — not just uncaught. Integrity by design, not by detection.

We go deeper on this in the AI Safety Center.