AI shame: why half your workforce is hiding their AI use, and what it's costing you

WalkMe's 2025 AI in the Workplace Survey of 1,000 US workers found that 49% of employees have concealed their AI use at work to avoid judgement, and 53% of C-suite leaders admit to doing the same. The people rolling out the AI strategy are quietly ashamed of using the tools they're telling everyone else to adopt. Alongside it, 78% of employees are using AI tools their employer never provided or approved: when people feel too ashamed to use the official tools openly, they turn to shadow AI instead, creating security and governance blind spots no dashboard is built to see.

This isn't just a feelings problem, it's measurable. A study published in Harvard Business Review in August 2025, led by researchers at Peking University and Hong Kong Polytechnic University, had 1,026 managers review identical Python code from a pool of 28,698 software engineers, varying only whether the engineer was described as having used AI assistance. Engineers described as AI users received 9% lower competence ratings for identical work, and the penalty wasn't distributed equally: a 13% drop for women against 6% for men, with senior male engineers who hadn't adopted AI themselves penalising female AI users up to 26% harder. A separate Duke University study in the Proceedings of the National Academy of Sciences, across four experiments with 4,439 participants, found the same pattern: people who use AI are perceived as lazier and less competent than people helped by a non-AI source, for identical outcomes, and managers who don't use AI themselves were significantly less likely to hire candidates who do. The Peking University study's sharpest finding: anticipated competence penalty predicted AI refusal more strongly than lack of training, age, or access to tools.

Most organisations measure the wrong things as a result. They track tool activation rates, time saved, prompts logged, and report all of it up the chain with confidence. What they don't measure is how many people feel incompetent using the tool daily, or how many are quietly opting out because it's safer than failing publicly. I see the same shape in almost every engagement: a company invests millions in new tools, training rolls out, and six months later actual usage is sitting at 15 to 20 percent because nobody addressed the emotional layer underneath the technical one.

AI Shame thrives specifically in blame cultures, where AI use becomes something to hide rather than something normalised and taught with clear guardrails. There's a perverse incentive that makes it worse, too: employees whose AI-driven productivity gains translate into higher expectations rather than more development time have a second reason to conceal what they're doing, since the honest answer invites more work rather than recognition.

What actually breaks the cycle isn't more training, since you can't train someone out of an emotional barrier. It's leaders going first with their own messy, imperfect AI use rather than a polished version of it, naming the emotional exhaustion of relearning a skill out loud, reframing AI use as a professional skill rather than a shortcut, building peer learning alongside formal training, and evaluating the quality of the work delivered rather than how it was produced. Performance systems that implicitly penalise visible AI use, especially for women and older workers who the research shows face harsher penalties, punish exactly the behaviour leadership says it wants.