Last week, a New York Times investigation found something that should stop every school administrator mid-scroll: students are handing AI agents the login to their entire online course — quizzes, discussion posts, essays, all of it — and the agent just… does the class. Nobody caught it. Not because the school wasn’t looking, but because there was nothing left to look at. The work got done. It just wasn’t done by anyone learning.
The response since then tells you more than the original story did.
Blackboard’s chief of staff, Jason Weaver, put it plainly: "there has never been more pressure on academic integrity." So Blackboard did something about it — it brought Cursive, a writing-verification startup, onto the team. Cursive doesn’t grade the finished essay. It watches the typing. Keystroke patterns, revision history, time on task — a fingerprint of the writing process itself, built to prove a human was actually there.
Meanwhile, educators asked the AI companies for something simpler: just have your agents identify themselves when they’re inside a learning platform, so a teacher can see what’s a student and what isn’t. Perplexity said no. Letting its agents self-identify, the company said, "would put the student’s privacy and security at risk."
Sit with that for a second. The companies whose products are completing students’ coursework are citing student privacy as the reason not to make that visible.
None of this is actually new, either. Vanderbilt disabled Turnitin’s AI detector back in 2023, once the math on its own claimed 1% false-positive rate worked out: with 75,000 papers a year, that’s roughly 750 students a year potentially told they cheated when they didn’t. Vanderbilt’s not alone — versions of that decision have been made across higher ed for three years running, and non-native English writers have been shown, repeatedly, to get flagged at far higher rates than everyone else. Three years later, the industry’s answer to a harder version of the same problem is a keystroke tracker.
That’s the pattern worth naming: every response to "how do we know a student actually did the work" has been a race to build a better lie detector. Detectors. Typing-pattern trackers. Refusals dressed up as privacy protections. All of it aimed at catching the fake after the fact — and all of it, so far, either wrong too often to trust or beaten within a semester.
Here’s what none of these approaches ask: what if you didn’t need to detect anything, because you were there for the real thing the whole way through?
That’s the actual difference between a verification tool and a facilitation model. A typing-pattern tracker tells you a human’s fingers were on the keyboard. It tells you nothing about whether a human was thinking. Auxesis’s Facilitation Brief exists for the opposite reason — it’s built around a real person watching a student reason through a problem in real time, in a live session, leaving a record of the actual thinking, not just the keystrokes that produced a document afterward. You don’t have to reverse-engineer whether learning happened. You were in the room for it.
The schools and families racing to build better detectors are solving the wrong problem. The question was never "how do we catch students who didn’t learn." It’s "how do we make sure they do." One of those is a surveillance project. The other is just teaching, done in a way that leaves proof behind.
Which would actually reassure you more: a tool that tracks your kid’s typing, or a teacher who was in the room the whole time?
Leave a Reply