"Admitting to Using ChatGPT Can Backfire: Why Employees Are Hiding Their AI Use"

Biases That AI Users Are "Less Competent and Diligent"
57% "Submitted AI-Generated Work as Their Own"
Growing Need for Clear Workplace Guidelines

The rapid spread of generative artificial intelligence (AI) use in the workplace has brought about unforeseen issues. While work efficiency has improved, distrust is growing among colleagues over who wrote reports or emails and how they were written. In addition, many employees are found to be hiding the fact that they use AI at work, fearing that admitting it might lead to being perceived as "lazy" or "lacking ability."

On September 11 (local time), The Wall Street Journal (WSJ) reported that "AI is creating new trust issues among coworkers," and warned that if companies fail to set clear guidelines for AI use, AI could end up undermining collaboration within organizations, despite its introduction for the purpose of improving productivity.

The rapid spread of generative artificial intelligence (AI) use in the workplace has led to unexpected issues. While work efficiency has improved, distrust among colleagues is growing over who and how reports or emails were created. The photo is an AI-generated image to aid in understanding the article. ChatGPT

According to a "Trust, Attitude, and Use of AI" survey conducted late last year by KPMG and the University of Melbourne on 48,340 people across 47 countries, 58% of workers said they deliberately use AI in their work, and 31% used AI every day or every week. In the survey, 57% of workers said they had either hidden their use of AI or presented AI-generated outputs as their own work. Around half of the employees also indicated that they had used AI in ways that violated company policy. In contrast, only 40% said their workplace had policies or guidelines for using generative AI, and just 47% had received AI-related training.

This tendency to hide AI use is rooted in the still-prevalent negative perception of AI. There is concern that using AI for work could result in being labeled a "lazy employee" or deemed someone who lacks expertise. Research has shown that these concerns are not unfounded. In a study published last year in the Proceedings of the National Academy of Sciences (PNAS) by researchers at Duke University in the United States, four experiments involving more than 4,400 participants demonstrated a “social AI penalty,” where people using AI tools were evaluated more negatively in terms of competency and work motivation than those who did not use such tools.

Participants in the experiment expected employees who used generative AI to be lazier, more replaceable, and to have less ability and diligence compared to those using standard work tools. They were also less willing to disclose AI use to their supervisors or colleagues. There is even an "AI disclosure paradox," where being transparent about AI use can actually lead to worse evaluations.

In a study published last year in the academic journal "Organizational Behavior and Human Decision Processes," 13 separate experiments confirmed that people who disclosed their use of AI received lower trust ratings than those who did not. Another study released this year observed an "AI penalty," where authors of workplace emails or social media posts were deemed less trustworthy and less authentic if it was known that AI had contributed to the writing.

The challenge is that humans are, in fact, not very good at accurately determining whether someone else has used AI. Citing relevant research, the WSJ reported that, generally, people could only identify AI-generated writing from text with 50 to 54% accuracy. In a 2025 study involving German university instructors who attempted to distinguish AI-generated research paper sentences, the proportion of AI-generated sentences correctly identified by humans as such was only about half. Another study involving prospective teachers also found that the overall accuracy for distinguishing AI writing was just 54.3%.

This suggests that if one jumps to the conclusion that a sentence was written by AI just because it is excessively smooth or formal, there is a risk of unfairly assessing the work capabilities of colleagues or subordinates, regardless of whether AI was actually used. Since employees may hide their AI use to avoid negative evaluations, while the lack of disclosure fosters suspicion among coworkers, a vicious cycle of mistrust may result.

Experts emphasize that, to address these problems, companies must do more than merely allow or prohibit AI; they should create detailed standards for allowable scope and responsibility according to job function. A study by IBM researchers this year also found that, compared to simply indicating "AI was used," specifying whether company AI policies were followed, which part of the work AI was used for, and whether a human reviewed the final result, all helped reduce the burden and negative perceptions of disclosing AI use. The WSJ also added that companies should anonymously monitor actual patterns of employee AI use, clearly define which use cases are permitted for each job, and continually share this information with all organization members.

This content was produced with the assistance of AI translation services.