Does hidden AI use cost more trust when exposed?

Schilke and Reimann's second claim is that hiding AI use is riskier than disclosing it, once others find out. The excerpt says: "If you're using AI on the job, the cover-up may be worse than the crime. We found that quietly using AI can trigger the steepest decline in trust if others uncover it later. So being upfront may ultimately be a better policy." The authors present this as a finding in its own right, separate from the disclosure penalty, and they frame the choice as a dilemma: users can "embrace transparency and risk a backlash, or stay silent and risk being exposed later – an outcome our findings suggest erodes trust even more."

The excerpt gives no mechanism for this finding. The reasoning it offers, that people still expect human effort, is attached to the disclosure penalty and is not extended to concealment. It also does not say what "steepest" is measured against: it names no comparison condition, no size of the decline, and no indication of whether this result comes from the same 13 experiments as the disclosure penalty. The phrase "even more" is the authors' comparison, made without figures in the passage.

The finding sits on top of the disclosure penalty in Does disclosing AI use damage how trustworthy you seem?, which is why the two should be read together. Disclosure costs trust; silence avoids that cost in the short run but adds a larger one if the hidden use surfaces. Does revealing AI identity help or hurt user trust? describes a parallel failure of concealment, through a different route. In that study, hidden identity produces no learning at all, because selectors "cannot attribute outcomes to partner type." Together the two notes describe concealment failing in two ways: it blocks the calibration that could repair disclosure's cost, and when the hidden use is discovered, it costs trust sharply. The settings differ, since the earlier study concerns partner choice in a hybrid society and this excerpt concerns workplace self-report, so the link is between parallel mechanisms, not between one shared result.

What the excerpt does not establish is how often hidden AI use is discovered, or whether exposure costs more than disclosure in absolute terms. The authors assert the comparison without reporting figures, and they do not say whether the trust loss from exposure recovers over time. The excerpt also does not test any of the three policy options it lists, so "may ultimately be a better policy" is a tentative implication, not a tested recommendation. The reading the evidence supports is narrower than a verdict for upfront disclosure: in the settings the authors studied, discovery is a real cost of silence, and that cost makes the choice between disclosing and concealing less one-sided than the disclosure penalty alone suggests. It does not establish that disclosure wins overall.