A 31-year-old physician was using DoximityGPT — HIPAA-compliant and organization-approved — with a personal prompting template he said produced “astoundingly good results.” His colleagues were struggling with the same tool, and he believed the template would help. He still didn’t share it.
The anecdote lands because it is so banal: nothing clandestine, nothing technically fancy, just a private workflow advantage sitting in someone’s drafts folder.
The scale of this silence is not small. A global KPMG and University of Melbourne study of more than 48,000 respondents found 57% of employees admitted to hiding their AI use at work. An Anthropic study found 69% of professionals mentioned social stigma around using AI at work. A separate survey of 604 U.S. employees who use AI daily or multiple times per day found 30.3% had intentionally withheld AI-related knowledge: workflows, techniques, prompts.
The deciding factor wasn’t simply whether a company had an AI policy or sanctioned tools. Neither predicted hiding on its own. Trust and psychological safety mattered more. Employees in the lowest trust quartile (measured through agreement with statements like “In general, I believe my employer’s motives and intentions are good” and “My employer is not always honest and truthful”) were nearly four times as likely to withhold AI knowledge as those in the highest: 47% vs. 14%. Psychological safety showed a similar split: 45% vs. 17%.
AI changes the economics of “organizational silence.” It’s no longer just hiding problems. It’s hiding solutions — like collapsing a three-hour task into 20 minutes with prompt sequences and chained tools that are easy to conceal and even easier to take with you.
Interviews pointed to three rational motives for staying quiet: reputational risk, because AI use can be framed as incompetence; workload “taxation,” where time saved on A and B becomes D, E, and F; and replaceability, because enterprise tools can log prompts and workflows, turning hard-won methods into transferable process maps. Texas A&M business school professor Matthew Call has warned that knowledge once accumulated through years of experience can now be extracted, stored, and handed to a replacement.
The kicker: approved enterprise tools amplify the relationship between trust and hiding. Where trust is high, people hide less. Where trust is low, they hide more. Logging can look like credit assignment or surveillance, depending on how much employees believe leadership is “foaming at the mouth trying to find ways to use AI to fire everyone.”
Full story at Harvard Business Review.