Higher education is quietly changing direction. Institutions that once viewed AI detection software as the solution to academic dishonesty are now beginning to step away from it. According to RTL Today, reporting on AFP, Yale and Cornell have banned or advised faculty against treating detection tools as primary evidence of cheating.
This retreat does not mean that AI use is being accepted without question. Rather, it reflects an acknowledgment that the evidence is unreliable. A detection score represents a probability, not proof. When a student’s degree or admission is at stake, that distinction becomes critical. Alternative assessment methods, such as oral examinations, also have limitations. They can favor more extroverted students while putting those who think more slowly and reflectively at a disadvantage.
Markus Goldbach has spent 15 years working in academic integrity technology. He is the CEO of Mentafy, a platform built around a different idea: the writing process itself provides the evidence. Goldbach believes the industry has been focusing on the wrong question. He explains, “AI detection software only evaluates silent grammar patterns and word frequencies. It produces a probability, a likelihood. There is no underlying proof whatsoever.”
These technical limitations are not simply a matter of debate. TBS Graduates, citing AFP, reported that the University of Wisconsin-Madison warns in its faculty guidelines that detection tools are imperfect at best, can produce false positives, have demonstrated bias against non-native English speakers, and will not stop students from using these tools.
Universities are already changing their approach. According to Inside Higher Ed, an increasing number of institutions have prohibited AI detectors, citing their unreliability as indicators of cheating. Yale, Vanderbilt, Johns Hopkins, and Indiana, for example, have introduced policies that ban or discourage relying on detection tools as sole evidence. At least a dozen universities have also disabled Turnitin’s detection software.
Goldbach explains, “Detection tools look for patterns in grammar, word choices, and style that AI tends to produce. But a student who happens to write in a similar style gets flagged, even if they never touched AI.”
An IEEE Xplore benchmarking study published in June 2026 evaluated five widely used detectors across 180 academic writing samples. The findings showed false positive rates ranging from 15.6% to 30.6%. Accuracy fell considerably when analysing hybrid texts combining human and AI writing, dropping to between 54.2% and 70.8%. When detection tools cannot consistently identify the work they are supposed to catch, they are no longer a reliable safeguard. They become a coin toss with serious consequences.
Goldbach adds, “Honest students are getting caught in the middle. They are accused of something they did not do, and they have no way to prove otherwise. The false negative rate is also substantial, and we’re not even accounting for that here.”
The solution is not to develop increasingly sophisticated detection software. Instead, institutions need to change what they examine. Mentafy records the drafting process from the initial blank page, following revisions, edits, and the development of ideas. Its purpose is to make AI use visible rather than hidden. A student using AI to polish grammar, a practice already permitted by most universities, leaves a different record from someone who generates an entire essay from a prompt. The writing process helps reveal the difference.

Goldbach is not opposed to AI. He compares its appropriate use to asking a parent or peer to proofread an assignment or asking a librarian to recommend a useful source. The difficulty is that detection tools cannot tell the difference between these permitted uses and generating an entire paper. When a polished sentence triggers a warning, they can penalize students who followed the rules while overlooking those who did not. ABC reported that hundreds of students at an Australian university were wrongly accused of submitting AI-generated papers. After months of waiting, the institution eventually determined that they were innocent. The technology has not become more dependable since.
The structural change Goldbach advocates is documentation. Universities should make digital records of the writing process part of academic work from the beginning, not as a form of surveillance but as evidence of intellectual effort. “The final PDF is not the learning,” he says. “The learning is the revision, the false starts, the moment a student realizes an argument does not hold.” This is the record that detection software cannot see, but educators need.
“If you do not have blood, sweat, and tears during your learning process, you will not learn that much. Your brain is a muscle; if you let AI do the heavy lifting, it will never get strong,” Goldbach says. His warning is more than a figure of speech. Meaningful academic work develops problem-solving stamina that no detection score can measure.
The longer-term consequences of doing nothing are significant. If degrees become little more than rubber-stamped AI prompts, employers may eventually stop placing value on them. Goldbach warns that the damage may not take five years to become visible. It could emerge much sooner, as a generation of graduates enters the workforce without the ability to think independently. For countries without natural resources, such as Germany, losing the capacity to innovate represents an economic risk as well as an educational one.
Moving forward does not mean returning to handwritten examinations. Oral assessments have a legitimate role, but they also carry their own biases, potentially favoring extroverted students and disadvantaging those who need more time to think deeply. The better approach is to retain long-form academic work while making the process behind it visible. Universities do not need to fear AI. They need to stop expecting algorithms to resolve a challenge that requires a fundamental change in how learning is assessed.
The true value of education was never just the final document. It has always been the intellectual effort required to produce it. Institutions that recognize this will be better positioned to navigate the AI transition while maintaining their credibility. Those that continue chasing detection scores risk contributing to the decline of an entire generation of graduates.


