Shadow AI: why your employees have stopped telling you about their AI usage
Shadow AI is not a discipline problem. It's a rational response to an incentive most leaders built by accident. Research shows over half of employees hide their real AI use, and a separate study shows exactly why: honesty about it gets punished. Banning tools and tightening policy makes the hiding worse, not better. The only way through is to stop asking people to confess and start measuring the work itself, which is the whole premise behind AI Litmus.
The gap
Two numbers, from two separate 2025 studies, describe the exact same blind spot.
The first number is from a global study of 48,000 workers across 47 countries, run by the University of Melbourne with KPMG. The second is from Slingshot's 2025 Digital Work Trends Report, a separate survey that found 45 percent of employees keep their AI use private to some extent. Different studies, different people surveyed, same story: employers consistently overestimate how much their teams are telling them.
If you run a team, the odds say your own read on your company's AI use is backwards.
It has a name: shadow AI
This behavior now has a name in the research: shadow AI, use that happens off the record, outside any tool the company sanctioned or any conversation the company can see.
It is not a fringe habit. It is the default response to a specific, well-documented incentive, and once you see the incentive, the behavior stops looking irrational.
Why hiding it is the smart move
Atlassian's Teamwork Lab ran a controlled test this spring. Nearly a thousand employees judged the exact same piece of work; the only thing that changed was whether the writer had disclosed using AI.
When honesty about AI gets you labeled lazy, the rational response is to stop being honest about it.
The one exception: companies whose culture already treats AI use as normal. There, the penalty nearly vanishes, and people who disclose are rated as more efficient, not less.
What silence actually costs you
A hidden behavior is not a controlled one. Your best people, the ones furthest ahead, have the most reason to stay quiet, so you learn nothing from exactly the employees you most need to learn from.
- You cannot spot who is genuinely good at this and scale what they do.
- You cannot catch risky use, sensitive data pasted into the wrong tool, before it becomes an incident.
- You cannot tell a team that isn't using AI apart from a team that uses it constantly and has just learned not to mention it. Two different problems, and you can no longer see which one you have.
Why the obvious fix backfires
Tighter policy, more monitoring, a longer ban list, all of it raises the cost of being honest even further. A rule that punishes admitting something does not shrink the something. It just shrinks the admitting.
This is a culture problem wearing a policy costume, and no monitoring software fixes a culture problem.
Stop asking. Start measuring.
If most employees will not tell you the truth about their AI use, asking was never going to work. A survey runs into the same wall a manager's gut instinct does.
This is exactly the gap AI Litmus is built to close: not by asking anyone to confess, but by reading the actual work, whether the outcome got better, faster, and more trustworthy on the tasks a role actually involves. You cannot out-survey a 57 percent concealment rate. You can only measure past it.
See this on your own teams.
A private walkthrough, calibrated to your roles. About two weeks.
Frequently asked
What percentage of employees actually hide their AI use from their employer?
A 2025 global study of nearly 48,000 workers across 47 countries, led by the University of Melbourne with KPMG, found that 57 percent of employees say they hide their AI use from managers and present AI-generated work as their own. A separate 2025 survey, Slingshot's Digital Work Trends Report, found 60 percent of employers believe their employees are being fully transparent about their AI use, while 45 percent of employees said they keep it private. Two different surveys, same gap between what leaders believe and what their teams actually do.
Won't banning unauthorized AI tools fix the problem?
The evidence points the other way. A controlled experiment by Atlassian's Teamwork Lab in 2026 found that workers who disclosed using AI were judged far more harshly than peers doing identical work who said nothing. A ban or a stricter policy raises the cost of being honest about AI use even further, which gives employees more reason to hide it, not less. The research found the stigma nearly disappears in companies whose culture already treats AI use as normal, which is a culture fix, not a policy fix.
If more than half of employees won't admit their real AI use, how can you possibly measure it?
Not by asking. Self-report and manager perception both depend on people telling the truth about something the research shows most will not disclose honestly. The alternative is to look at the work itself: whether the actual outcome on a role's real tasks got better, faster, and more trustworthy, which does not require anyone to confess anything.
Shobhit Khandelwal is the founder of VMS Culture Labs, on a mission to measure what most leaders only guess at: how fluently their teams truly work with AI, and the hidden cost of how people behave at work. He is out to replace workplace guesswork with evidence, and build the kind of workplaces the next generation deserves.
Connect on LinkedInOccasional, high-signal notes on measuring AI fluency and culture cost. No spam.