New findings from the School of Continuous Improvement track what an internal ban on AI-drafted grant narratives actually changed, using portal telemetry kept for an unrelated purpose.
A telemetry feed kept for an unrelated reason finds the drafting moved off the portal rather than stopping
An attestation checkbox is hard to fail. That may be the least surprising finding in a new paper from the School of Continuous Improvement, which set out to check what changed after Slop University banned AI-drafted first drafts of internal grant-application narratives, and found the checkbox itself the only thing that moved on schedule.
The team set the ban’s own compliance record against telemetry the University’s grants portal has quietly kept for years, for an unrelated reason, across 349 grant narratives submitted over the six months either side of the ban’s start date, then checked the pattern against a round of staff interviews. Attestation compliance reached full completion within the first reporting cycle. The telemetry record moved only at the margins, and mostly in narratives that stopped showing any drafting activity on the portal at all, rather than in narratives that kept drafting there.
None of this works without a telemetry feed the portal has quietly kept for years for an entirely different reason. We didn’t build anything new; we just finally asked it something.
— Associate Professor Casimir Beng, Lead of the Adaptive Metrics Lab
“Everyone we spoke to already knew exactly why they had moved the drafting off the portal,” said Dr Anneke Tolan, Senior Research Fellow at the School of Emergent Priorities. “It wasn’t hidden from us; it just isn’t something the audit log was ever going to ask about.”
The School describes the divergence as reason enough to watch its own compliance systems as closely as anyone else’s.
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.7aynun.