Half of Workers Hide Their AI Use Despite Leadership Encouragement

The hidden cost of mixed messages

More than half of employees who use AI tools at work won't admit to using them for important tasks, according to research from Microsoft and LinkedIn. The data reveals a troubling pattern: 78 percent of AI users bring their own unauthorized tools to work, 52 percent hesitate to disclose AI use on critical projects, and 53 percent worry that admitting they use AI makes them appear replaceable.

The contradiction stems from leadership behavior. Executives publicly encourage AI experimentation while simultaneously praising work done "without AI" or making jokes about AI-assisted emails. These mixed signals teach employees that AI adoption carries social and professional risk despite official policy.

Dr. Gleb Tsipursky, CEO of Disaster Avoidance Experts and author of The Psychology of AI Adoption at Work: From Resistance to Results, devotes a section of his new book to this dynamic. He argues that shame-driven secrecy around AI use creates what he calls "shadow AI"—unauthorized tool adoption that bypasses security review and compliance oversight.

Why it matters

When employees hide AI use, organizations lose visibility into which tools touch sensitive data, how decisions get made, and where accountability lies. Shadow AI creates the same risks as shadow IT but moves faster and touches more workflows. Leaders who want the productivity benefits of AI need employees to disclose usage—but disclosure requires removing the social penalty that currently attaches to it.

Four practices that reduce shadow AI

Dr. Tsipursky outlines practical steps leaders can take to make AI use transparent without creating stigma.

Define acceptable use explicitly. Most companies tell employees to use AI "responsibly" without defining what that means. Workers need clear categories: approved uses, approved uses requiring disclosure, restricted uses needing permission, and prohibited uses. The NIST AI risk management framework provides a foundation for building these guidelines.

IBM Research found that granular disclosure reduces perceived stigma when employees explain how AI contributed, confirm they followed policy, and state that a human reviewed the output. A useful disclosure might read: "AI generated the first draft. I checked the facts, revised the reasoning, and approved the final version."

Model transparent disclosure from the top. The reluctance to admit AI use extends beyond early adopters. Pew Research Center found that 52 percent of U.S. workers feel worried about future workplace AI use, and 33 percent feel overwhelmed. Harvard Business Review has documented an "AI penalty" that can attach to workers when colleagues know they used AI assistance.

Leaders who want employees to disclose AI use must disclose their own first, demonstrating that transparency carries no professional cost.

Separate tool approval from individual judgment. Organizations need processes for vetting AI tools separately from evaluating whether an employee made a good decision. When these blur together, employees learn to hide tool use to avoid scrutiny.

Reward disclosure, not concealment. Recognition systems that celebrate "human-only" work or joke about AI assistance teach employees that disclosure is risky. Leaders should instead acknowledge employees who follow disclosure protocols and use AI within approved boundaries.

The details were first reported by Inc., drawing on Dr. Tsipursky's research and data from Microsoft, LinkedIn, Pew Research Center, IBM Research, and Harvard Business Review.