Employees hide their AI use when getting better only earns them a lower rate or more work. It's a rational response to a ratchet, not a character flaw. The cost to you is twofold: risk you can't see, and productivity gains that never spread past the person who found them.
This post is about the incentive side of hidden AI use, written for owners and executives. (The security side, where IT has to stop being the department of no, is in our post on shadow AI.) The numbers say this is common. A 2025 global study of more than 48,000 people by KPMG and the University of Melbourne found that 57% of employees say they hide their use of AI and present AI-generated work as their own.
Why do employees hide their AI use?
Start with a scene most people know, even if they've never watched the episode.
In a 1952 episode of I Love Lucy called "Job Switching," Lucy and Ethel take jobs at a candy factory, wrapping chocolates as they come down a conveyor belt. Their supervisor warns them that if a single piece gets past them unwrapped, they're fired. The belt speeds up. They can't keep pace, so they start hiding chocolates: in their mouths, in their blouses, in Lucy's hat. When the supervisor comes back, she sees no unwrapped candy and concludes they're doing fine. Her response is to call out to speed the belt up.
That's the ratchet. Show that you can handle more, and you get more. So the smart move is to never let anyone see you have room.
Ethan Mollick, a Wharton professor, has a name for the AI version. In a 2023 post, he called them secret cyborgs: employees who use AI to do their work better, often discovering genuinely valuable methods, and keep it hidden from their employers. Secret cyborgs are Lucy with the chocolates in her hat.
More recent survey data points the same way. In a 2026 PagerDuty survey of office professionals, 39% said they would rather use AI without telling anyone, and the most common reason given for hiding it, at 33%, was to avoid scrutiny from managers or leadership.
What are the three ratchets?
Hidden AI use usually traces back to one of three ratchets. Most organizations have at least one without meaning to.
The rate ratchet
When pay or pricing is tied to output, getting faster can lower what each unit is worth. Hit the target too easily and the target moves. Anyone who has worked under a quota knows how this goes: you learn to deliver just enough.
The load ratchet
Same pay, more work. Someone uses AI to finish their job in less time, and the reward is a second job's worth of tasks. Or a vacancy goes unfilled because "you're managing fine." The person who found the gain ends up carrying the cost of it.
The billable-hour ratchet
This one matters most for law firms, accounting firms and anyone else who bills by the hour. If AI turns a five-hour task into a one-hour task, the client pays for one hour. Unless the firm changes how it prices or measures people, the person who found the faster method has just lowered their own billable total, and possibly the firm's revenue. In a business built on hours, nobody should be surprised that efficiency stays quiet.
What does hidden AI use cost an organization?
Hidden AI use costs a business in two ways, and both get worse the longer it goes on. The first is risk. When people hide their AI use, nobody checks which tools hold client or company data, whether personal accounts are involved, or whether anyone reviewed the output before it reached a client. The second cost is quieter and often larger: the gains don't spread. The employee who figured out how to cut a weekly report from three hours to thirty minutes keeps that method to themselves, so the other ten people doing similar reports never learn it. The organization pays full price for work that could be faster, while the improvement sits in one person's private routine. Monitoring doesn't fix either cost, because the cause is an incentive problem. People hide gains when showing them would cost them something. The fix is to change what showing a gain earns.
How do you get employees to share their AI gains?
The short version: don't take the gain. Buy it.
If the organization simply claims every hour someone saves, people will stop saving hours in public. If it pays for those hours, with a share of the benefit and a promise about what improvement won't cost, people will bring their methods forward. Four moves make that real.
1. Promise, in writing, what a gain won't cost
The promise has to come from whoever controls budget and headcount. A manager's email that says "we value innovation" doesn't count. Something like "no one's pay, role or billable expectations will be reduced because they found a faster way to do their work this year" does.
2. Set the bar from outside the person
Base expectations on the role, the market or the team, never on an individual's own best quarter. If one great month becomes the new floor, everyone learns to avoid great months.
3. Share gains by a formula published in advance
Expectations will eventually change; that's how businesses improve. When they do, share what the change saves or earns according to a rule everyone could read beforehand: time back, a share of the savings, a bigger role, recognition. The specific formula matters less than the fact that it's public and decided before anyone has something to lose.
4. Make it safe to show
Create a simple, low-stakes way for people to show what they've built. Celebrate the first few who do. Don't respond to the first disclosure with an audit.
How can you tell whether your organization has a ratchet?
Try an exercise we call trace the gain. List the last five improvements anyone in your organization made, with or without AI. For each one, write down where the saved time went.
If most of the answers are "more work for the same person," "a position we didn't refill" or "a lower price for the client with no change for the team," you have a ratchet, and your people already know it. If some answers are "time back," "a bigger role" or "a share of the savings," you're building the opposite.
People change when they see what's in it for them. If your change doesn't show it, change the change, not the message.
Isn't this just an AI policy problem?
Partly. Your policy should make the rules about data and approved tools clear, so nobody hides because they're unsure what's allowed. But a policy can't promise what a gain won't cost. That's a leadership commitment, and it's the part of AI strategy most organizations skip. We separate the two in AI policy vs. AI strategy.
If you lead a firm in Cedar Rapids, Iowa City or elsewhere in Iowa and you suspect your best people are wrapping chocolates out of sight, our AI leadership and strategy work starts with finding where the gains are going. Learn more on our Advise, Educate & Lead page.