The Hidden Assist

In short: Many people hide their AI use at work because they fear looking less capable or replaceable. That hiding breaks quality control, stops shared learning and leaves risk unmanaged. Organizations should make honest disclosure the safest choice instead of trying to police use.

Almost everyone I work with uses AI. Far fewer talk about it.

They'll mention it in passing, a little sheepishly. "I had it clean up the email." "I used it to get a first draft going." Or they won't mention it at all, and you only find out when a document arrives with a phrase nobody on the team would ever say. A 2024 survey by Microsoft and LinkedIn found that roughly half of people using AI at work were reluctant to admit using it for their most important tasks. That matches what I see.

The instinct is understandable. The consequences are worse than most leaders realize.

Why Do People Hide Their AI Use?

The reasons are human and predictable.

Fear of looking less capable. If the work came from a tool, does that mean you couldn't have done it yourself? Many people worry their credibility depends on the work looking unassisted.

Fear of looking replaceable. If a model can do part of your job, maybe someone will decide it can do all of it.

Unclear rules. In many organizations nobody has said what's allowed. When the rules are vague, hiding feels safer than asking.

Mixed signals from leadership. Leaders praise AI adoption in all-hands meetings and then raise an eyebrow at work that "sounds like ChatGPT." People learn quickly that the safe move is to use it and not say so.

What Does Hidden AI Use Cost?

In my thesis on using AI, I argued that the real value of these tools is making reasoning visible. Hidden use does the opposite. It makes reasoning invisible at exactly the moment it matters most.

Quality control breaks. If a reviewer doesn't know a section was generated, they don't know to check it differently. AI errors look different from human errors. They're confident, plausible and often wrong in specific, checkable ways. A reviewer who knows to look will catch them. A reviewer who doesn't, won't.

Learning stops. When people hide what works, nobody else learns from it. The person who figured out a great way to use AI for client research keeps it to themselves, and the team reinvents it badly ten times over.

Risk goes unmanaged. Hidden use means hidden data exposure, hidden errors and hidden dependencies. An organization can't govern what it can't see.

Trust erodes in both directions. People feel like they're getting away with something. Leaders suspect everything is generated. Neither side believes the other is being straight.

Why Don't Strict AI Policies Work?

The natural response is to tighten the rules. Ban certain uses. Require approval. Make examples of people caught breaking policy.

This usually makes the problem worse. Strict rules don't stop people from using tools that make their work easier. They just drive the use further underground. You end up with the same amount of AI use and less visibility than before.

The organizations that get this right do the opposite. They make honest disclosure the safest choice available.

Policing by detection doesn't work either. AI-detection tools are unreliable: they flag human writing as generated and miss generated text that has been lightly edited. An organization that relies on them will punish some honest people and miss most of the actual use.

How Do You Make AI Disclosure Safe?

Say clearly what's allowed. Spell out what AI can be used for, what needs review and what's off limits. Ambiguity is what makes hiding feel necessary.

Judge the work, not the tool. If the work is good, accurate and thoughtfully reviewed, it shouldn't matter how much a model helped. Evaluate the outcome and the judgment behind it.

Make disclosure routine and boring. A one-line note on significant work: "Drafted with AI, reviewed and edited by me, figures checked against source." When everyone does it, nobody feels exposed by doing it.

Leaders go first. The fastest way to make disclosure normal is for senior people to talk openly about how they use AI, including where it helped and where it got things wrong.

Share what works. Turn good uses into shared practices. Celebrate the person who found a better workflow instead of quietly wondering if they cut corners.

Should You Tell Clients You Used AI?

Inside an organization, the case for openness is simple. With clients it takes more care. Some contracts have rules about AI use or confidentiality. Some clients are paying for a particular person's judgment and reasonably want to know how it was produced. And anything pasted into a tool may leave the building, so what goes in matters as much as what comes out.

The principle is the same, scaled to the stakes: be clear about how AI fits into the work, never put confidential information into tools that aren't approved for it, and don't let a client believe something is entirely hand-made when it isn't. Clients rarely object to AI. They object to finding out later.

Honesty as Infrastructure

It's tempting to see AI disclosure as an ethics issue, a matter of personal integrity. It is partly that. But in organizations it's mostly an infrastructure issue. Honest disclosure is what lets teams review work properly, learn from each other, manage risk and trust what they're reading.

The question isn't whether your people are using AI. They are. The question is whether they feel safe telling you how. If they don't, you've lost the visibility you need to use these tools well, and you've traded it for the comfortable fiction that the work is all human.

I'd rather have the truth. Most leaders would, if they thought about what the fiction is costing them.

Frequently Asked Questions

How many people hide their AI use at work? A 2024 Microsoft and LinkedIn survey found roughly half of people using AI at work were reluctant to admit using it for their most important tasks.

Do AI detectors work? Not reliably. They flag some human writing as generated and miss generated text that has been lightly edited.

What should an AI disclosure note say? One line is enough: what the tool did, what the person did and what was checked. For example: "Drafted with AI, reviewed and edited by me, figures checked against source."

Should agencies tell clients they use AI? Be clear about how AI fits into the work, follow any contract rules, never put confidential information into unapproved tools and don't present AI-assisted work as entirely hand-made.

Evan Prout is a strategist in Grand Rapids, Michigan, who writes about how organizations think and decide with AI.