A new survey of 7,200 adults across the United States, Europe, the United Kingdom, and Latin America has laid bare a striking contradiction in attitudes toward artificial intelligence in the workplace. Forty-six percent of respondents said they would publish AI-generated work without disclosing that AI played a role, even though 54 percent said other people should disclose AI use for equivalent tasks. The findings, released in late July, suggest that personal behavior and public expectations are diverging faster than workplace policies can adapt.
The survey was conducted by an AI aggregation platform that offers a single interface for multiple AI models, giving the company a direct commercial stake in normalizing AI adoption. Its managing director argued that the results demonstrate people already see themselves as legitimate authors when they guide AI to produce work. The platform's press release did not disclose the sampling method or margin of error, making it difficult to assess how representative the results are. But the findings align with a growing body of academic and industry research on AI concealment, ownership, and stigma.
The ownership paradox
The survey probed when people feel entitled to call AI output their own. Sixty-nine percent said the work belongs to them if they supplied the instructions and approved the final version. Seventy-four percent felt the same after revising AI-generated material. Seventy-one percent said taking responsibility for accuracy was enough to claim authorship. These numbers indicate a broad psychological shift: many workers no longer view AI as a mere tool but as an extension of their own creative process, akin to a spell-checker or a grammar assistant.
That sense of ownership, however, does not translate into transparency. The same respondents who claim authorship are often unwilling to tell their managers, clients, or collaborators that AI was involved. The gap between what people expect from others and what they practice themselves has been documented repeatedly. A separate survey by ResumeBuilder found that nearly 63 percent of workers have never told their managers that AI was doing part of their job. Half of Gen Z workers report feeling guilty about using AI, even as employers increasingly rank AI skills above a university degree in hiring decisions.
Fear of professional consequences
The reasons for hiding AI use are becoming clearer. A June analysis published in Harvard Business Review concluded that employees hide AI usage largely because they fear professional consequences, not because they lack awareness of disclosure norms. Workers worry about being perceived as lazy, incompetent, or dishonest. Those fears are not unfounded. Atlassian, a software company that has studied collaboration patterns, found that workers who disclose AI use are judged ten times lazier by colleagues. This perception penalty can affect performance reviews, promotions, and everyday working relationships.
The stigma is particularly acute in fields where individual expertise and originality are prized, such as journalism, law, academia, and consulting. In these environments, admitting that AI helped draft a report or analyze data can undermine professional credibility. Yet the same workers may expect others to be transparent. This hypocrisy is not necessarily deliberate; it reflects the absence of clear social norms. When there is no agreed-upon etiquette for AI disclosure, people default to self-protection.
Ownership and accountability in the age of AI
The survey's findings on ownership raise important questions about accountability. If a worker supplies detailed prompts, iterates on AI output, revises the text, and takes responsibility for its accuracy, many would argue that the work is genuinely theirs. But what about workers who use AI with minimal oversight and then present the result as their own? The line between assistance and substitution is increasingly blurred.
Legal frameworks have not kept pace. In most jurisdictions, copyright law grants protection only to human-authored works. The U.S. Copyright Office has repeatedly ruled that purely AI-generated images and texts without sufficient human involvement are not eligible for copyright. The European Union's AI Act, which began imposing new transparency obligations on August 2, takes a different approach: it focuses on labeling rather than authorship. Under the AI Act, certain AI-generated content that could be mistaken for authentic must be clearly labeled. The regulation target deepfakes and synthetic media, not workplace documents, but it signals a broader regulatory shift toward mandatory disclosure.
Companies are beginning to respond. Some have introduced internal policies requiring employees to flag AI-assisted work. Others have banned generative AI tools altogether, fearing data leaks and legal liability. Many are still in a gray zone, leaving individual workers to decide whether to declare their AI use. The result is a patchwork of norms that varies by industry, region, and even team.
The broader context of AI adoption
These disclosure dilemmas are emerging against a backdrop of rapid enterprise AI adoption. According to a McKinsey survey, nearly three-quarters of organizations have adopted AI in at least one business function, and a significant share uses generative AI tools such as chatbots and image generators. Employees are often ahead of their employers in exploring these tools, using personal accounts to circumvent corporate oversight.
The disconnect between private behavior and public expectation is consistent across age groups and geographies. Older workers tend to be more cautious about AI disclosure, while younger workers are more likely to feel guilty about using AI. But the overall pattern is the same: people expect transparency from others but hesitate to offer it themselves. This is a classic collective action problem, where individual incentives conflict with the group's stated values.
Some researchers argue that the focus on disclosure is misplaced. They suggest that what matters is not whether AI was used, but how it was used and who is responsible for the final output. A surgeon who uses AI to analyze medical images is still accountable for the diagnosis. A reporter who uses AI to transcribe interviews is still accountable for the quotes. In this view, ownership should be determined by human oversight and accountability, not by the presence or absence of AI assistance.
Workplace policies and the road ahead
Most workplaces have not yet developed clear rules on AI disclosure. A 2024 survey by the Society for Human Resource Management found that fewer than one-third of employers had policies specifically addressing generative AI. Those that do exist often focus on data security and plagiarism, rather than on when and how to disclose AI use to colleagues or clients.
Human resources professionals are now grappling with a new set of questions. Should employees be required to mark AI-generated paragraphs in reports? Should managers disclose when they use AI to evaluate employees? Should AI use be included in performance reviews as a competency or as a risk? These questions have no easy answers, and the lack of consensus is visible in everyday practice.
At conferences and industry events, executives pay lip service to transparency while admitting privately that they do not always follow their own rules. A technology manager at a marketing firm told researchers that she uses AI for brainstorming and first drafts, but she rarely tells her boss because “he would think I'm not doing my job.” Another employee at a financial services company said he stopped using AI altogether after a colleague was quietly reprimanded for using ChatGPT on a quarterly report.
The EU AI Act and the trajectory of regulation
Regulation is starting to catch up with these ambivalent attitudes. The EU AI Act's transparency provisions took effect on August 2, 2024, requiring providers and deployers of certain AI systems to ensure that content generated or manipulated by AI is labeled when it could be mistaken for authentic. While the immediate focus is on deepfakes and synthetic media, the legislation sets an important precedent: the public has a right to know when they are interacting with AI-generated material.
Businesses outside the EU are also taking note. Multinational companies are being forced to comply with the EU rules for their European operations, and many are extending those requirements globally to avoid inconsistency. This regulatory drift toward disclosure is expected to continue. Lawmakers in the United States, Canada, and the United Kingdom have all proposed or debated AI transparency legislation, though nothing as comprehensive as the EU's framework has passed yet.
For now, the gap between belief and behavior remains wide. People overwhelmingly think AI use should be disclosed, just not by them. Workplaces have yet to produce rules clear enough to close the distance. The survey's contradiction is not a failure of individual morality; it is a symptom of a society that has not yet decided what honesty means when machines participate in human creativity.
As AI tools become more integrated into daily professional life, the pressure to resolve this contradiction will only intensify. Employees want to use AI to improve their productivity and career prospects, but they also want to preserve their identity as skilled, autonomous workers. Employers want to encourage AI adoption to remain competitive, but they also want to manage risk and maintain trust with clients and regulators. Neither side has found a stable equilibrium. The result is a period of awkward silence, where AI use is simultaneously everywhere and nowhere.
The survey's numbers may vary across subgroups, but the fundamental tension is universal. People are more willing to forgive their own AI use than others'. They are quick to claim ownership when it suits them and quick to demand disclosure when it does not. Until organizations establish clear, practical, and non-punitive guidelines for AI transparency, the contradiction will persist. The technology is advancing faster than the social contract, and no regulatory text has yet closed the gap.