Why employees don’t admit to using AI

Why admitting to using AI can feel risky

A specialist prepares a very good analysis, using AI along the way to organise the material. In a meeting, he is asked: “Did you write this yourself?” If the tone suggests that using a tool reduces the value of the work, next time he may not mention it. The problem need not be an intention to deceive the company. The employee may believe that his effort, knowledge and decisions will be ignored once he reveals that AI was involved. Where doing a task single-handedly is valued above all else, that fear is understandable.

Research on organisational silence has shown for years that people weigh up the risk of voicing information that could damage their standing. It would be wrong to conclude from this that everyone who stays silent is afraid of the boss. Some simply don’t know which uses need to be disclosed. So the first conversation should be about rules and reasons, not an attempt to unmask hidden users.

Research points to a fear of being judged as less competent

A 2026 study by Mizrak and colleagues covered 564 consulting and professional services employees in Turkey who had experience of using generative AI. The authors analysed the relationships between concealing the use of these tools and fear of disclosure, concern about losing one’s reputation for competence, and the clarity of company policies. Greater fear was associated with more self-reported concealment of AI use. This does not prove that these factors are the cause, nor is it a representative result for every profession.

There is an important difference between assessing the outcome of work and holding a stereotype about how that outcome was produced. If a manager praises a report but, on hearing that AI was used, immediately decides the author is less capable, they send a contradictory signal. People will look for ways to protect their image. The answer is not to excuse every use of AI unconditionally, but to set proportionate requirements and clear quality criteria that don’t depend on fashionable declarations.

Unclear rules create a grey zone

In one department, management encourages automation; in another, people still hear that “a real expert does everything themselves”. No one has explained whether client data may be entered into an external tool, whether AI can be used for the first draft of a contract or whether it can generate a suggested reply. Employees see a gap between the pressure to be productive and the risk of breaking the rules. Some give up on the technology; others start using it quietly. That is a problem with how work is organised, not just with attitudes.

Good rules should answer simple questions: which tools are approved, what data must not be entered, which uses require disclosure, who verifies the output and what to do if something goes wrong. Not everything needs the same level of reporting. Correcting the punctuation in an internal note and preparing a recommendation for a client based on confidential data are not equivalent activities.

Calling something “generated” does not remove accountability

Transparency does not mean adding “made with AI” to every sentence. Where a use has consequences for a client, a patient, a financial decision or the law, what matters is what the tool did, where the data came from and who assessed the result. The author of a report should be able to defend the sources used and understand the recommendation. Saying that “the model suggested it” is not enough. On the other hand, disclosing the use of AI should not diminish the value of work that has been done thoroughly.

A team can agree on three levels: language support, substantive support in creating content, and involvement in analysis that informs an important decision. Each level carries different requirements for checking and disclosure. The exact boundaries depend on company policy, client contracts and legal obligations. Employees should see not only the prohibitions, but also access to approved tools that let them do their work within the rules.

Ask about working practice, not about guilt

A bad way to open the conversation is: “Own up – who’s using AI and pulling the wool over my eyes?” It assumes guilt before the rules have even been explained. A better version: “I want to work out where AI genuinely helps, where mistakes creep in and which rules we need to make clearer. We won’t treat simply using an approved tool as a lack of competence.” A statement like that only becomes credible when the manager’s actual reactions live up to it.

In a one-to-one, ask: “At what stage of the work was the tool helpful? What did you check yourself? What do you need to do this safely?” Don’t probe for private details unrelated to the task. If confidentiality has been breached or sources misrepresented, assess the specific behaviour in line with company rules. Taking action over a specific mistake should not mean condemning every new solution.

A culture of openness needs predictable reactions from managers

Employees learn the rules not only from documents, but from how the first report of a problem is received. If a manager thanks someone for flagging an incorrect result and helps put it right, they show that problems can be talked about. If they mock someone who asks about an approved tool, they undermine their own message about openness. Research on psychological safety points to links with information sharing and team learning, although it does not promise a simple, guaranteed effect from a single intervention.

In practice, it is worth regularly collecting examples of successful AI use, but also of situations where verification uncovered a mistake. The aim is not a public list of culprits, but a shared understanding of the boundaries. Clear rules, good tools and a fair way of assessing work can reduce the need to hide anything. The manager then builds a team in which using technology is neither a secret nor an excuse to avoid responsibility.

Not disclosing the use of AI at work can stem from fear of being judged, unclear policy and mixed signals from management. That does not excuse breaching confidentiality or presenting unverified content as one’s own analysis. But it is worth separating those problems from the use of the tool itself. Transparent rules, proportionate oversight and a calm conversation about quality help turn suspicion into collaboration.

How Empatyzer helps you talk about AI without a witch-hunt

A manager suspects that some of his team are using AI outside the official rules, but he doesn’t know whether the reason is a lack of clear guidelines, awkward tools or fear of being judged. Instead of starting with accusations, he describes to Em what the current rules say and what behaviour he has actually noticed. Em helps him separate facts from assumptions, phrase neutral questions about working practice and prepare a short explanation of the rules in force. In the team context, Em can take into account differences in attitudes to new solutions and in the need for security. When a difficult conversation with a particular employee comes up, the available relationship tips help him choose a way of giving feedback without shaming anyone. The manager also asks Em for help in answering an employee’s question: “If I use AI, why do other people think I’m less competent?” The conversation ends with a plan: which cases require disclosure, how to document that the output has been checked and how to report a problem. Micro-lessons help reinforce the skills of giving constructive feedback and communicating technological change.