While corporate executives globally push for the swift adoption of generative artificial intelligence, daily office dynamics reveal a deep psychological barrier. A behavioral study conducted by the Hebrew University of Jerusalem and WHU Otto Beisheim School of Management demonstrates that using large language models at work carries tangible risks for individual career progression. Employees who utilize generative software and transparently report doing so face an immediate and substantial erosion of trust from direct supervisors.
The research team, consisting of Samuel Meyers, Dr. Yochanan Bigman, and Dr. Roman Briker, carried out six controlled experiments involving more than 1,000 managers for their study, which was published in The Leadership Quarterly. Participants evaluated identical workplace reports where only the declared method of preparation varied. When subordinates produced their work entirely by hand, managers trusted the submission without inspection in 68 percent of cases. When identical work was delivered with the help of generative artificial intelligence, that trust rate plummeted to 24 percent, regardless of the objective quality of the output.
This 44 percentage point drop in managerial confidence triggers severe downstream consequences for staff. The findings confirm that supervisors do not merely scrutinize the work more skeptically, but also rate the promotability and leadership potential of AI-using employees significantly lower. Subordinates who transparently admitted using software assistants received systematically worse performance ratings and were disadvantaged during bonus allocations. The researchers refer to this phenomenon as an institutional AI penalty.
Clear institutional governance serves as the primary mechanism to mitigate this dynamic, although it fails to erase it completely. The experiments show that explicit corporate permission or active encouragement by executive management raises managerial trust to 47 percent. While this marks a notable recovery compared to outright bans, where trust crashed to 19 percent, or environments lacking any policy at 33 percent, it still remains far below the 68 percent baseline established for unassisted human labor.
This divergence creates a difficult double bind for white-collar professionals. Workers face mounting executive demands to deliver higher output at accelerating speeds through modern automation tools. Concurrently, behavioral cues from line managers signal that leveraging such technology will be interpreted as a lack of effort, intellectual shortcutting, or diminished personal competence. Those who operate with full transparency risk undermining their standing during future appraisal rounds.
The findings emphasize that issuing basic technical guidance is insufficient to modernize corporate workflows. As long as managerial intuition equates unassisted manual labor with diligence and integrity, using generative assistance will remain a career hazard for individual workers. Organizations must rethink how they evaluate knowledge work, shifting focus away from perceived human labor time toward actual qualitative outcomes.