Generative AI has upended a core assumption of educational assessment: that the work a student submits can stand as reliable evidence of the human abilities a degree or certificate claims to represent. A new preprint on arXiv argues that the problem is not simply whether students use AI, but what the resulting work can certify at all.
Because AI can produce fluent, plausible output across many domains, the traditional link between performance and competence becomes fragile. The paper frames the challenge as one of 'cognitive stewardship'—deciding how to design and interpret assessments in an environment where AI participation is pervasive and often invisible.
Notably, the abstract suggests the issue is deeper than plagiarism detection. Even if AI use were perfectly identified, the question remains: what does the credential actually certify? The paper calls for rethinking assessment rather than merely policing tools.