Credential farming in the age of generative AI: relocating trust from outputs to communities

Publications, affiliations and authorship positions began as proxies for qualities that cannot be observed directly: judgement, understanding and the capacity to contribute to shared inquiry. Generative artificial intelligence has weakened the inference those proxies depend on. Polished research outputs can now be produced by people who do not understand them, yet admissions committees and hiring panels still read an output as evidence of the competence that producing it once required. The result is faster credential farming, the extraction of the social trust held in credentials without contribution to the enterprise that gives them meaning. Better detection of machine-generated text will not repair this. What has to change is where epistemic trust sits. Work in the sociology of credentials, in social epistemology and on situated learning points to where it should move: scientific formation, how communities shape people and teach them what to value, carries the evidential weight that credentials only summarised. Three commitments follow. Select members for depth rather than legibility. Build small relational structures in which judgement can be formed and witnessed. Refuse to convert those structures into new metrics. The argument matters most for researchers in low-resource settings, whose learning happens in informal spaces that credentialing systems do not see, and who are asked to make themselves legible in a vocabulary nobody taught them.