AI Detectors Are Out, New Approaches Are In

As faculty prepare their assessments for the new academic year, many are facing an artificial intelligence–powered academic integrity conundrum.

First, there’s the undeniable reality that access to AI tools have made large-scale cheating easier than ever. Autonomous AI agents have shown that they can complete—and ace—entire online courses without any help from the student enrolled in the class. Faculty at Brown and Alcorn State Universities went viral this summer after setting traps that indicated the majority of their students used AI to complete major assessments. And according to one recent national survey, 73 percent of faculty say they have personally dealt with academic integrity issues involving AI.

At the same time, it’s becoming increasingly difficult to hold students accountable for suspected unauthorized AI use. Numerous studies have shown that AI detectors—such as those developed by OpenAI, Writer, Copyleaks, GPTZero and CrossPlag—exhibit inconsistencies, produce false positives when evaluating human-generated writing and are biased against non-native English writers. In some cases, those false positives have resulted in accused students filing lawsuits against their universities.

That reality has prompted a growing number of institutions—including Yale, Vanderbilt, Johns Hopkins and Indiana Universities—to enact academic integrity policies that ban or discourage faculty reliance on AI-detection programs as the sole evidence of alleged cheating. At least a dozen universities, including Northwestern, Georgetown and New York Universities, have also disabled Turnitin’s AI detection software, which was found to have a much higher false-positive rate than the company originally suggested.

“Some tools claim to detect whether a student used AI, but these services are highly unreliable,” reads the newly updated Faculty AI Playbook at Indiana’s Kelley School of Business, which prohibits the use of AI detectors. “Instead of trying to ‘catch’ AI use, focus on designing assignments that encourage process, reasoning, and authentic engagement.”

For some universities, the turn away from AI detectors is also motivated by the fact that students are increasingly deploying AI to game the detectors.

“Evidence also shows that ‘humanizing’ programs are very successful at reducing or eliminating the percentage of AI usage that is successfully detected in generative AI-produced writing,” Jennifer Frederick, executive director of Yale’s Poorvu Center for Teaching and Learning and her colleague Alfred Guy, director of the center’s undergraduate writing and tutoring programs, wrote in a joint email to Inside Higher Ed. “We want to avoid the inevitable cat and mouse game created by AI detection tools: As the tools get better at ‘catching’ AI generated text, so do methods to evade detection. This becomes a technical exercise rather than a learning event.”

‘Difficult, Delicate Moment’

So where does this leave faculty members who want to ensure that students are learning and not offloading thinking to a machine?

“It’s a difficult, delicate moment right now. And I’m not sure we have all the answers,” Kevin Yee, director of the Faculty Center for Teaching & Learning at the University of Central Florida, told Inside Higher Ed. “If faculty had an AI detector that was reliable, that’s what they would use. But many are reluctantly moving in the direction of assignment redesign because they are at wit’s end” over AI-powered cheating.

From what he’s observed, about half of faculty want to make as few changes as possible to AI-proof their assessments. But staying the course may be a viable option only for those who teach in-person classes with fewer than 30 students. In a small environment, “they can actually succeed in convincing students not to take these shortcuts, establishing strong relationships and convincing students that the learning is really what matters here,” he said. “In larger classes, there’s less of a personal connection between students and faculty, so faculty need to redesign their assessments to be more AI-resilient or a completely kind of deliverable that AI can’t do, such as a blue book exam or an in-person oral exam.”

To help faculty redesign their assignments, Yee co-authored a guide that contains dozens of suggestions centered on making students show how they are using AI to support their work. Faculty can also use AI to facilitate reworking their assessments by feeding the assignment into a large language model and asking for help making it more AI-resilient. “It doesn’t have to be a multi-week process,” Yee said. “AI can provide different ideas that don’t make course redesign nearly as hard.”

Although it creates additional work for faculty, redesigning assessments that can’t be outsourced to AI is critical for institutions on the hook to prove the value of their course offerings. “This is going to take eternal vigilance,” he said. “AI is never going away and it’s going to proliferate into alternative formats, and many endanger the educational experience. Thinking and learning are meant to be effortful and AI reduces effort and friction.”

But getting the majority of faculty on board with changing traditional assessments will be “the biggest lift” for universities responding to rising AI-related academic integrity concerns, said Marc Watkins, a writing and composition lecturer and director of the AI Institute for Teachers at the University of Mississippi.

“I can’t even fathom the cost this will have for a single university, let alone the majority of our campuses, to go through this,” he told Inside Higher Ed, citing the human labor and resources required to scale AI-resilience tactics, such as proctoring oral exams. No matter the challenges, though, “It shouldn’t all be on faculty to [navigate]. They should be asking for support.”

‘The Real Trick’

Securing a commitment from faculty to overhaul courses will also require some deep self-reflection from academics themselves, said Tricia Bertram Gallant, director of the academic integrity office and testing center at UC San Diego and co-author of The Opposite of Cheating: Teaching for Integrity in the Age of AI.

“Over the past 20-plus years, there have been changes that have slowly degraded the validity of our 20th-century assessments,” she said, pointing to the rise of internet-supported plagiarism, the contract cheating industry and now AI. “Yet, for the most part, higher education hasn’t changed. We’re still relying on the unsupervised written word as evidence of learning. The real trick is acknowledging that that doesn’t work anymore.”

And now that AI has made the temptation for students to cheat on traditional writing assignments so strong, Bertram Gallant said it’s no longer feasible for faculty to foster academic integrity by policing AI use.

“We cannot be giving unsupervised assessments to students expecting them to resist AI. They’re not going to be learning, we’re going to be spending our time trying to catch them cheating and they’re going to graduate and realize they just wasted four years,” she said. Instead, she advises departments to focus on teaching the foundational skills required to use AI responsibly and ethically. “We need to give students a distraction-free, temptation-free environment to tell us what they know and what they can do.”