409: LMS Says It Cannot Help You Stop AI Agents
Plus, the Linkletter/Proctorio story finally ends. Plus, cheating in Texas. Plus, where was I?
Subscribe below to join 5,010 (-1) other smart people who get “The Cheat Sheet.” New Issues every Tuesday and Thursday.
The Cheat Sheet is free. Although, patronage through paid subscriptions is what makes this newsletter possible. Individual subscriptions start at $8 a month ($80 annual), and institutional or corporate subscriptions are $250 a year, suggested. You can also support The Cheat Sheet by giving through Patreon.
LMS Says It Can’t Detect Or Stop AI Agents
This is important, I think.
Over the past few months, a healthy number of people have become worried about AI agents — the bots that can complete an entire learning module, or entire online course, by themselves, unassisted by any human, in minutes. Tell it to take the course and pass the test, and it does. Simple as that.
If you’re unfamiliar with AI agents, “agentic AI,” here is a video you should see:
Setting aside that for some reason this is the threat that has people up in arms — it was just fine, I guess, when a student had to do the hard work of asking Chegg for the answers — agentic AI is a serious problem for academic integrity.
Spotted by a reader, the news here is that Anthology (Blackboard) one of the larger LMS providers said a few weeks ago that they cannot stop AI agents from obliterating their assessments:
AI Agents don’t show up as separate apps or users—they look identical to normal student activity in the LMS. Because they operate at the web interaction and automation layers—filling in forms, clicking buttons, making API calls, etc.—they fall outside what the LMS itself can control. Many are completely invisible to the platform.
And:
In short, 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.
First, I am not entirely sure I believe this is true — that it is “not possible” to block AI agents (see Issue 405). As I said there, the bigger question to me is whether the LMS companies want to block them. I think they don’t.
The bigger issue here, however, is that LMS is kicking this problem, as it has with every other integrity-related issue, back to schools and teachers. Anthology says:
any short-term responses to the potential misuse of AI Agents will need to be delivered through institutional policies.
I highlight “policies.” As opposed to practices. Anthology goes on:
Institutions can also guide students through updating their generative AI policies to clearly define the permitted and prohibited use of AI Agents and providing thorough training on these updated policies.
This is, on its face, useless. Updating your policies and telling students not to do this is no solution. I do not know whether Anthology is really this clueless, but it’s not a great look.
Consider also, from Anthology, in the same article:
Here at Anthology, our fundamental view remains—as we’ve articulated previously in the context of AI plagiarism—that blocking AI in the classroom is extremely difficult, and institutions will see best results by adapting their pedagogical approach to the AI era.
Don’t try to block AI, in other words, change your teaching. Thanks, Anthology.
But there is more, as Anthology adds:
Rethinking assessment practices emerges as a key priority in this context.
Change your teaching and assessments. Change your policies. Don’t bother to try to police the AI use that shortcuts actual learning because it’s “extremely difficult.”
There’s no question that dealing with AI use on assessments is, in fact, extremely difficult when your LMS cannot, or will not, help.
Imagine that someone builds and sells you a house. You can do anything and everything in it, they say. It’s full of features and safe. Then it comes out that some new technology allows someone to take control of your house — know when you’re home, open and unlock doors, remove the roof, burst pipes, whatever they want.
Then imagine you go to the guys who built the house and sold you on it, and you say, “Hey, this is a really big problem.” To which they reply, “Sure is. Nothing we can do about it. We suggest you make it clear to the people taking the roof off of your house that this is not allowed. And also, you know, you should rethink what it means to have a home. Also, good luck.”
That, here, is Anthology. And be honest, in this analogy, would you at least stop paying for this house? Demand a refund? What good is a house if it fails the basic deliverables of a home? What good is an LMS — a learning management system — if it cannot manage learning?
At the end of this, AI — including agentic AI — is a problem. And your LMS provider has shown they are not the least bit interested in helping you solve it. Keep paying those fees though. They are very interested in that.
Cheating in Texas
A few weeks ago, The Houston Chronicle ran a pretty lengthy story (subscription required) about AI use and academic misconduct. It starts with a student at University of Houston who:
In English and government, she must stick to approved tools and stay within the school’s 20% threshold, measured by special AI detection software. But her history professor bans AI entirely. There, Romero can’t touch it — or she’ll risk violating UH’s academic honesty policy.
So?
It is, of course, mind-bendingly dumb, in my view, to set an AI-use threshold and tell students what it is. May as well just give them the University President’s ATM PIN code.
Student: AI, write this for me. Also student: AI, use a humanizer to make it less likely to be flagged by an AI detector. Also student: Let me check the AI detector. Also student: Let me revise it or humanize it again to get my score down below 20%.
Also student: I am so confused about AI policies that I cannot know what to do.
The news article continues:
The AI boom has turned college classrooms into a patchwork of policies, with many professors setting their own rules for everything from multiple-choice homework to essays — leaving both students and instructors unsure where appropriate AI use ends and cheating begins.
Not buying it.
Students know what cheating is. They know when they’re taking shortcuts. And, if they don’t, they can — wait for it — ask. Crazy, I know. I’ll even bet these policies are in the course syllabus.
Aside, I am always baffled as to why we seem to insist that the brightest, most promising, most educated young people in the world are stupefied by situation-dependent rules. You can drive 60 mph on most highways, but not in residential school zones; check the signs. You can text your friends all you want at the baseball game, not in the movie theater.
No one can be expected to manage their lives in such a “patchwork of policies.”
Moving on:
“It can almost create a culture of paranoia for students who are living in constant fear of being called out for possible AI use, when they’re trying their best not to,” said Lauren Zentz, who chairs the UH English department and reviews academic integrity cases. “It’s just a little bit of a minefield.”
Nonsense. Absolute nonsense.
“They’re trying their best not to [use AI]?” What? Who is making students use AI? Especially when they’ve been informed not to? Is that really a valid defense at UH? Sorry, I mean, I tried my best not to turn in the work I did not do, but it’s so complicated. And I’m so worried about, “in constant fear” of, being accused of using AI that I just had to use AI. I tried so hard not to.
This is not a credible thing to say.
Later, the paper says:
Still, some students cheat — and professors say those cases have made things harder for everyone.
True. Some students cheat. Lots and lots of students, if you believe the data. And that does indeed make things harder for everyone. Good catch right there, major American newspaper. You cannot buy analysis and insight like that.
The story states again that AI use policies “vary widely” and how:
The result can be confusing for students
Still not buying it.
But here’s a sensible quote from the story:
“The reason we leave it up to the faculty is because it’s really for them to know what are the learning outcomes,” Rice University Provost Amy Dittmar said. “This is a tool that, if used properly, may help you towards those learning outcomes, if not used properly, could hinder you towards those learning outcomes.”
Imagine that. Maybe — just maybe — trusting the people we’re paying to help students learn should decide what learning looks like in their fields, courses, and cohorts. But, for some reason, this is presented as a problem. That reason is that it’s confusing — for college students.
Not to put too fine a point on this, but were I ever in a role in which I had to interview recent college graduates for career opportunities, I’d ask if the patchwork of rules around using AI was confusing. If they said yes, I’d seriously question their ability to do any job well.
Anyway, the piece continues with a professor who has returned to hand-written blue books for in-class assessments:
[The professor] said his return to blue books happened after a disastrous experience teaching one asynchronous course called “Film and Existentialism.”
He found that students who couldn’t pass the multiple-choice quizzes would submit “thoughtful, seamless” papers.
“I knew it simply wasn’t them,” he said. “I learned two lessons: never to try asking for papers written again, and the second is never to teach an asynchronous class.”
No surprise whatsoever.
This professor goes on to say that he cannot trust Turnitin’s AI detection due to false positives:
He and several other professors said in multiple instances Turnitin over-detected students’ AI use in their classes, leading to allegations of academic dishonesty that cost students time and peace of mind — and could result in a failing grade or even expulsion.
Zentz, the professor who said that students were trying so darn hard to not use AI said:
in one case, a student wrote a paper with a tutor at UH’s writing center, submitted it and then received a warning that the work was 100% AI.
Not sure I buy that one. But whatever.
More:
Some Texas universities have chosen to avoid the problems that might arise from AI detection software. UT-Austin doesn’t have a contract with any, partly because they haven’t been proven effective, according to Julie Schell, UT-Austin’s assistant vice provost of academic technology.
I’ve written a few times about what a ridiculous thing this is — see Issue 232 or Issue 296. Speaking before of hiring recent college graduates, a degree from Texas at Austin would absolutely get my attention right now because I could not be sure the degree-holder did any of the work to earn it. Neither can the school.
I’ll end with this, from the story:
several professors have become adept at detecting AI themselves, spotting its tone and patterns. And they’ve gotten crafty to catch AI cheaters.
Gonzalez said he’s heard of professors hiding instructions written in invisible ink in their assignments, so students who generate their essays with AI end up with nonsense words in their copy. Some of Gonzalez’s instructors have also stopped uploading professionally-developed instructional materials onto Canvas and draw problems by hand instead, so AI can’t read them.
“Professors are getting smarter,” Gonzalez said.
Proctoring Critic and Proctorio Settle Legal Cases
The strange and occasionally sad story of Ian Linkletter, the former university staffer whose public criticisms of assessment proctoring led to him being slapped with an injunction from Proctorio, is finally over. The parties settled.
I’m not going to rehash the story, but bits of it are in Issue 103, and Issue 206, and Issue 90, and Issue 267.
The headlines here are that, in response to the injunction suit from Proctorio, Linkletter counter-sued, claiming he was the victim of legal harassment designed to silence him. For a hot minute, he was the martyr of the day for anti-proctoring folks, raising money for his legal case and finding his way into countless news stories. Linkletter never won a single hearing on his case.
In the settlement, news outlets reported that no money was exchanged and that the injunction Proctorio originally sought remains in place. Predictably, Linkletter called it a win. According to the coverage, he wrote on his blog:
“I’ve won my life back!” He stated.
“It has been 1,899 days, each one of them felt,” Linkletter wrote. “My patience, determination, and resilience were tested, but I persisted, and an immeasurable weight has now been lifted.”
A true hero for academic fraud, that guy.
Whatever. It’s over.
Where Did I Go?
Some readers may have noticed that the last Issue of The Cheat Sheet came out two weeks ago — that we’ve been absent from your inboxes for a fortnight. What happened?
Well, two weeks ago Tuesday, I was traveling to the OLC Conference in Florida where I was giving a presentation on assessment security. I’m important like that. Then, two weeks ago Thursday, I was giving my presentation at said conference. See, important. Then last Tuesday, I was not feeling 100% and very behind on actual work due to twice-mentioned conference. And last Thursday was Thanksgiving, an important US Holiday. So, two weeks. Poof.
If you missed The Cheat Sheet, which I doubt, I am sorry. But I am back now. Sorry.
The Cheat Sheet is Verified as Human-Written.
