What are employees hiding when they use AI at work?

A team lead reads a client report that is cleaner than anything the analyst has written before. Nobody mentions how it was drafted. The lead does not ask, and the report goes out on Friday.

A global survey of more than 48,000 people in 47 countries, run by the University of Melbourne with KPMG, found that 57 percent of employees say they hide their use of AI and present AI-generated work as their own. Almost half admit to using AI in ways that contravene company policy, including uploading sensitive company information into free public tools. Many rely on AI output without evaluating its accuracy (66 percent), and 56 percent say they have made mistakes in their work because of AI.

Why people hide it

Hiding is a rational response to the signals many workplaces send. Only 47 percent of employees in the study say they have received AI training, and only 40 percent say their workplace has a policy or guidance on generative AI use. Half say they worry about being left behind if they do not use AI. Put those together and you get people under pressure to use a tool, with no rules for using it and no reason to believe that admitting it is safe.

A ban pushes the use further into personal accounts, where you have even less visibility. A reviewer who does not know AI drafted the work will not look for the errors AI tends to make.

Making disclosure ordinary

The goal is to make "I used AI for this" a boring sentence. Publish a short list of what may go into which tool, in plain language, on one page. Then ask for disclosure in the place the work already moves, such as a line on the client deliverable checklist or a note in the ticket. Disclosure should lead to a check on the work, and the person who disclosed should see nothing worse than that.

Leaders have to go first. When a manager tells the team which part of their own report a model drafted and how they checked it, the team learns that disclosure is safe. A policy document alone rarely teaches that.

Training closes the rest of the gap. The 66 percent figure is a skills problem as much as a culture problem, because people who do not know where a model fails will not check the right things. Our AI training workshop covers guardrails for data and accuracy on the team's own tasks. On the design side, who is the user when an AI tool never leaves the building? asks whether an employee can refuse a bad suggestion without leaving the task.

What to check this quarter

1. Ask staff anonymously which AI tools they use for work, including personal accounts, and compare the answer with the list you think you have.
2. Publish a one-page permission list and keep it current.
3. Add a disclosure line to the workflows where client or customer work leaves the building.
4. Ask each manager to disclose their own use first, in a meeting, with the check they ran.

If the anonymous answers show far more use than your licence count, you know where to start.