AI-Generated Work Is Breaking the Links Between Degrees and Real Skills, Researchers Warn

The rapid spread of generative artificial intelligence into universities is no longer just a problem of cheating students and alarmed professors. According to a new theoretical paper published in the journal Higher Education, it has become a systemic crisis that strikes at the very foundation of what a degree is supposed to mean. The study, authored by Yuke Meng, Xufeng Zhang and Han Li, introduces a concept the researchers call Synthetic Credentialism: a condition in which AI-generated or AI-assisted academic outputs allow students to obtain credentials that no longer reliably signal the competences those credentials were designed to certify. In other words, the diploma itself may survive, but the information it carries about a graduate’s abilities may quietly decay.

The authors build their argument on a well-established body of economic and sociological theory. Since Michael Spence’s classic 1973 model of job market signaling, economists have understood education as a mechanism through which individuals communicate otherwise unobservable qualities to employers. A degree functions as a costly, hard-to-fake signal of competence. Signaling theory, screening theory and the economics of information asymmetry, famously illustrated by George Akerlof’s 1970 market for lemons, all rest on the assumption that the signal is trustworthy. Sociologists such as Randall Collins and Ronald Dore added a critical dimension, arguing that credentials also serve as instruments of social stratification and that societies can develop a pathological appetite for qualifications in themselves, the so-called diploma disease. Synthetic Credentialism, the researchers argue, describes what happens when generative AI contaminates the signal at its source.

The framework developed in the paper has three dimensions. The first is Signal Decoupling: the growing gap between the output a student submits and the underlying competence that output is supposed to represent. When an essay, a code assignment or a literature review can be produced by a large language model in seconds, the artifact no longer guarantees that its author possesses the analytical, writing or technical skills it displays. The decoupling is not merely a matter of dishonesty; it is a structural property of assessment formats that assume authorship equals ability.

The second dimension is what the authors call Verification Collapse, defined as a threshold failure of institutional attribution. Universities certify competence by attributing submitted work to the student who submits it. That attribution depends on verification mechanisms: supervised examinations, plagiarism detection, viva-style questioning, writing-style comparisons. The paper argues that generative AI can push these mechanisms past a tipping point at which institutions can no longer credibly distinguish human-produced from machine-produced work. Empirical findings cited in the study underscore the point. Research on AI-text detection tools has shown both unreliable performance and systematic bias, with one widely discussed study in the journal Patterns finding that GPT detectors disproportionately flag writing by non-native English speakers. A major 2023 testing study published in the International Journal for Educational Integrity likewise concluded that detection tools cannot yet be trusted to adjudicate academic integrity cases.

The third dimension is Normative Ambiguity. Unlike contract cheating, where a student pays a third party and nearly everyone agrees the behavior is fraudulent, AI assistance occupies a gray zone. Students may use chatbots to brainstorm, outline, edit or even draft work in ways that institutional policies struggle to classify. Because the norms themselves are unsettled, responsibility becomes diffuse, and the traditional misconduct-based response, investigation and punishment of individual offenders, loses analytical traction. The authors deliberately shift the lens away from individual wrongdoing and toward the institutional conditions under which verification fails.

One of the paper’s most striking claims concerns where the damage concentrates. The forms of assessment most central to higher education, the take-home essay, the written assignment, the problem set, the reflective portfolio, are precisely the formats most vulnerable to AI substitution. Meanwhile, the formats that resist AI most effectively, such as supervised examinations, oral defenses and performance assessments, tend to be marginal in many programs, expensive to administer at scale, or pedagogically limited. This asymmetry means that risk is not distributed evenly across the system: it pools at the core. The essay, which for centuries has served as both a learning tool and a certificate of thought, is the artifact most directly hollowed out.

The paper also examines how vulnerability varies across disciplines. In computer science, empirical studies have shown that AI code generators can solve a large share of introductory programming exercises, undermining a staple of technical assessment. In law, experiments such as the well-known ChatGPT goes to law school study demonstrated that language models can pass components of professional examinations. In medicine and the health professions, where assessment of clinical judgment and professionalism carries life-safety stakes, the challenge is compounded by the difficulty of scaling authentic, human-supervised evaluation. Disciplines built on extended written argumentation face perhaps the deepest exposure, because writing has historically been inseparable from both learning and certification.

Beyond the classroom, the authors trace the equity consequences of credential erosion. If AI access is unevenly distributed, students with better tools, subscriptions and digital literacy can convert machine assistance into grades more efficiently than disadvantaged peers, importing new inequalities into the certification process. Paradoxically, the same technology can also disadvantage students who write nothing wrong: as detection tools misfire, non-native English writers have been falsely accused, and honest students face a presumption of suspicion. At the labor market end, employers who lose faith in degrees may demand alternative signals, portfolios, proctored tests, micro-credentials, shifting the burden of proof onto all graduates, including those whose competence is genuine. Credentialism, once criticized for inflating the value of paper qualifications, may now face the opposite pathology: deflation of the signal itself.

Why, the authors ask, do institutions find it so hard to respond? The answer lies in what they describe as a response dilemma rooted in institutional sociology. Drawing on the classic account of institutionalized organizations as structures held together by myth and ceremony, the paper argues that universities face pressure to maintain the appearance of reliable certification even when their verification capacity erodes. Tightening assessment through surveillance and detection risks false accusations, legal exposure and reputational damage, and the evidence suggests current detectors cannot bear that weight. Redesigning assessment toward authenticity, process-based evaluation and supervised performance is costly, slow and pedagogically demanding. Doing nothing risks a slow collapse of trust. Each path carries institutional costs, and the path of least resistance, continuing as before while formally banning AI, preserves legitimacy in name while allowing decoupling to deepen in practice.

The paper is careful to insist that the story is not one of inevitable ruin. Its final analytical move identifies assessment redesign as the principal route to restoring credential reliability. The authors draw on decades of assessment scholarship, including work on authentic assessment, evaluative judgment and sustainable assessment, to argue that the response must be architectural rather than punitive. Assessments that require students to show their reasoning, defend choices in interaction, connect work to lived contexts and demonstrate judgment over time are harder to synthesize and more informative even when AI tools are legitimately used. The deeper contribution of the framework, however, is diagnostic. By naming Synthetic Credentialism and decomposing it into signal decoupling, verification collapse and normative ambiguity, the study gives researchers, institutions and policymakers a shared vocabulary for a problem that has so far been discussed mostly in fragments. The question it leaves open is how quickly universities can rebuild attribution mechanisms before the credential they certify stops meaning what the world still assumes it means.

Subject of Research: A theoretical framework for how AI-generated competence signals undermine academic certification in higher education

Article Title: Toward a framework of “Synthetic Credentialism”: AI-Generated competence signals and the crisis of academic certification

Article References: Toward a framework of “Synthetic Credentialism”: AI-Generated competence signals and the crisis of academic certification. (n.d.). https://doi.org/10.1007/s10734-026-01773-4

Image Credits: AI Generated

DOI: 10.1007/s10734-026-01773-4

Keywords: Synthetic Credentialism, generative artificial intelligence, higher education, academic credentials, signalling theory, assessment redesign, academic integrity, verification collapse, credentialism, large language models, AI detection tools, equity in education