Why do knowledge workers hide signs of using GenAI?

Semi-structured interviews with 19 knowledge workers — recruited across IT, Data Science, Law, Management/HR, Creative industries, and R&D via Prolific, each using GenAI at least weekly for work, conducted December 4, 2024 to January 17, 2025 — find that "the ability to remove cues indicating GenAI use was perceived as validation of domain expertise." Colleague knowledge-sharing helped participants learn GenAI competencies, but the same participants valued erasing the "tells" of GenAI output in their own work. One participant, P14, "utilised GenAI to solve a colleague's problem, but took personal credit for developing the solution."

The paper sets this against prior literature that reads GenAI-hiding as rooted in "shame" or as a bid to "elevate one's status." The authors found both motives present but add a third: successfully erasing GenAI tells can be "a positive indication of one's professional expertise," independent of stigma, because it demonstrates the user didn't need to lean on the tool, or could finish what the tool produced to a standard that no longer shows its origin. The behavior lets a worker look more capable than colleagues who "fail" to hide their use. The cost is structural: erasing the cues that would let a colleague recognize GenAI involvement also removes their occasion to ask about it, which the authors say "reduced opportunities for learning via knowledge sharing" and fed "cultures that lack transparency" — worse still "in organisations with unclear policies."

This sharpens two related notes. Which workplace cues survive AI mediation and which disappear? names provenance as a cue workers actively protect because it stays "contestable" — tied to a recognizable source a colleague could question. This paper gives a second, more active reason to erase that same cue: doing so isn't only defensive impression management, it functions as a status move that credits the eraser with expertise. It also extends Do people fear judgment when they use AI at work?, which measured stigma as the reason stated disclosure drops. This study's interviews describe a motive beyond anticipated penalty — hiding as a positive expertise claim — and trace a specific downstream cost that the disclosure-intention study does not: concealment forecloses the informal, peer-to-peer channel these same workers rely on to build GenAI literacy outside formal training.

The sample is 19 Prolific-recruited participants from organizations with no explicit GenAI prohibition — the authors state plainly that the goal "is not to produce a generalisable account of human behaviour." The excerpt gives no count of how often hiding occurred, no behavioral check on what participants reported, and no measure of how much learning was actually lost. At the strength the evidence allows, the study supports a mechanism — expertise-signaling concealment competing with knowledge-sharing — observed in this sample, not a claim about its prevalence or scale across knowledge work generally; the clearest implication is that unclear organizational AI policy is where this tension is reported as sharpest, which points to policy clarity as a plausible lever rather than a demonstrated fix.