Reporters at The New York Times spent months looking into something that sounds almost too strange to be a trend: students handing their entire online course over to an AI agent. Not asking AI for help with an assignment. Not using it to summarize a reading. Logging an AI agent directly into the course platform — Canvas, Blackboard, whatever the school uses — and telling it: log in and complete my quiz. Watch the lectures. Take the tests. Write the papers. Even show up in the discussion board.
The student never opens the course. The AI does the whole thing.
This is worth sitting with for a second, because it’s not really a cheating story. It’s a preview of what “AI replaces the student” looks like when nobody stopped it.
Why the usual response doesn’t work here
The instinct is to talk about detection — can the school catch this? The honest answer, per the investigation, is: not reliably. Officials at Blackboard’s parent company put it plainly: “given currently available technologies, it is not possible for Blackboard — or any other LMS vendor or provider of a web-based service — to reliably detect an AI Agent, much less block one.” Some educators have asked the AI companies to make their agents identify themselves when they’re operating inside a course. Perplexity, one of the companies behind these agents, told the Times that doing so “would put the student’s privacy and security at risk” — and declined.
So detection isn’t coming to save this. And even if it did, catching the agent doesn’t fix the actual problem: a student who handed over an entire semester never learned the material the course was supposed to teach. Winning the arms race against the agent doesn’t produce a student who understands anything.
This is the mission filter, playing out at the extreme
Every piece of AI-in-education content we publish gets run through two questions: does this augment the student’s intelligence, or replace it? Does it deepen learning, or shortcut it? Most of the time, those questions are about degree — a little shortcutting here, a little augmentation there.
This story isn’t a matter of degree. It’s the replace end of the spectrum, complete. The AI isn’t augmenting anything. It’s standing in for the student, end to end, and the student comes out the other side with a grade and nothing else.
That’s useful, actually — not as an alarm bell, but as a landmark. It’s the clearest possible illustration of the thing we’ve been naming all along: the gap between looks completed and actually learned. A finished course with a passing grade tells you nothing about whether a single concept landed. This story just makes that gap impossible to miss, because there’s no ambiguity left — the student wasn’t even in the room.
What this actually means for the people who read this newsletter
If you’re homeschooling, this isn’t really about your kid logging an agent into Canvas — it’s the same warning in miniature. Any time “AI can technically finish this for me” is true of an assignment, some version of this failure mode is available, just smaller and less dramatic. The fix isn’t watching more closely. It’s designing the work so finishing it requires the thinking, not just the output — the same principle whether we’re talking about a full online course or a single math worksheet.
If you’re an educator, this is the strongest case yet for building courses and assessments around live explanation, iterative checking, and the kind of Socratic back-and-forth an agent can’t fake convincingly over time — not because agents can’t produce a plausible-sounding answer once, but because they can’t sustain a real, adaptive conversation about why the answer is right the way a student who actually did the work can.
The lesson from this story isn’t “ban agents” — that fight is already lost, per the people who study it for a living. The lesson is: any assignment an agent can complete without the student ever engaging is an assignment that was never actually testing learning in the first place. That was true before agents existed. This story just makes it undeniable.
If an assignment can be finished by an AI agent from start to finish, with the student never really engaging — was it ever actually testing learning?
Leave a Reply