Artificial Intelligence and the Risk of Deskilling in Conformity Assessment

Artificial intelligence (AI) tools are entering accredited conformity assessment, where the competence of assessors, auditors, inspectors, laboratory personnel and decision-makers is a constitutive element of confidence in the result. This article examines the risk of deskilling: deterioration, or insufficient maintenance, of task-specific independent professional capability through cognitive offloading, automation bias and overreliance. The argument rests on existing competence-maintenance requirements, domain-general evidence on offloading and automation bias, and experimental evidence on AI-assisted documentary reasoning and on skill formation when decision aids are withdrawn; clinical deskilling studies are treated as peripheral corroboration. The analysis examines the normative documents of the eleven Global ACI MRA scopes together with ISO/IEC 17011, ISO 19011, ISO/IEC 42001/42006, Regulation (EU) 2024/1689 as amended in 2026, and current accreditation-body guidance. Across these instruments competence, authorisation and accountability are assigned to persons and bodies, while AI is governed through validation, monitoring, security and human-oversight controls; none establishes AI as competent conformity-assessment personnel. Because many conformity-assessment judgements have no single external ground truth, erosion of independent reasoning may not be exposed by plausible output. The article argues that AI may substitute for tasks but cannot fulfil roles reserved to competent persons; distinguishes augmenting uses, through which accumulated expertise is exercised, from substituting uses, through which it is drawn down; proposes functional analogues of the established impartiality threats for human–machine interaction; and observes that interaction with a tool lacks the human friction through which competence is formed. Six proportionate safeguards and a research agenda are proposed. Direct longitudinal evidence in this population remains limited; the article therefore advances a risk-based competence-management argument, not a claim that deskilling has been demonstrated. The governing principle is that AI should augment professional competence, not become its source.