Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
From Forbid to Reimagine: Employer Strategies for Responding to Candidate GenAI Use in Assessment
Citation. Dunlop, P. D., & Lievens, F. (2026). From forbid to reimagine: Employer strategies for responding to candidate GenAI use in assessment. International Journal of Selection and Assessment, 34(2), e70060. https://doi.org/10.1111/ijsa.70060
Summary. This peer-reviewed commentary responds to prior discussion of candidate GenAI use in personnel selection and proposes the FAIR framework: Forbid, Advise, Insulate, and Reimagine. It argues that emerging agentic AI and AI-integrated wearables increase opportunities for GenAI to substitute for candidate effort, while unrestricted use can alter what assessment scores measure. The authors frame longer-term assessment design around fair, useful evaluation of human–AI collaboration rather than blanket prohibition.
Key claims
- Candidate GenAI use can threaten construct-related validity when scores reflect GenAI literacy and access in addition to the intended candidate attributes.
- Differences in candidates’ access to, willingness to use, and ability to use GenAI can create fairness and subgroup-difference concerns in selection outcomes.
- The FAIR framework distinguishes defensive responses (Forbid, Advise, Insulate) from Reimagine, which redesigns assessment around intended human–AI collaboration.
- Agentic AI and AI-enabled wearables may make conventional countermeasures, including digital proctoring and some face-to-face assessment practices, less reliable over time.
Evidence & limitations
- Commentary synthesizing author arguments and prior commentaries; it does not report a new empirical study.
- The article is an open-access, peer-reviewed journal commentary published online on 2026-03-30 and in volume 34, issue 2.
- The Drive PDF is the publisher version of record; extraction succeeded across five pages and no model-directed prompt-injection language was found after contextual inspection.