Transparency on AI, a risky bet for trust
Generative artificial intelligence is now making its way into every professional sphere, from writing simple emails to creating advertising campaigns, including human resources management, financial analysis, and even artistic design. Faced with this growing penetration, the question of transparency regarding the use of AI is becoming a burning issue: should we admit that we have used artificial intelligence to produce content, a report, or even to make a decision? This question, seemingly purely ethical or regulatory, actually conceals an unexpected paradox and a profound transformation in the dynamics of trust in the world of work and within the market.
Moral instinct and official recommendations are increasingly pushing for the disclosure of AI use. This transparency is presented as an ethical imperative, a demonstration of loyalty to employees, customers, students, or the general public. However, a much more nuanced reality is emerging as practices evolve and empirical research multiplies: declaring the use of AI, far from strengthening trust, tends to erode it, sometimes significantly and lastingly. This "transparency penalty" phenomenon is upending preconceived ideas about the link between openness and credibility, and shaking up the foundations of perceived legitimacy.
Across a series of experiments conducted in contexts as diverse as universities, businesses, recruitment, investment, and design, the finding is consistent: whether a professor, manager, creative, or company, anyone who admits to using AI to accomplish a task is deemed less trustworthy than someone who remains silent about this technological assistance. Even more surprising: this is not a one-off or anecdotal reaction, but a robust effect, across the nature of the tasks, the status of the beneficiary (student, employee, client, consumer), or the type of organization. Experiments also show that this loss of trust is not mitigated by the way the disclosure is formulated: whether it is specified that the AI was only used to proofread, to suggest ideas, or whether it is insisted on human proofreading, doubt about the legitimacy of the result sets in, and trust decreases.
This phenomenon is rooted in the perception of legitimacy. In the collective imagination, the normative expectation remains strong: a job, a decision, a production must be the result of human expertise, judgment, and creativity. As soon as the presence of AI is revealed, adherence to these standards crumbles: the product or decision is then perceived as “less human,” and therefore less trustworthy, even if the AI has objectively improved the process. Perceived legitimacy collapses, and trust evaporates, because the use of AI appears as a deviance from the implicit standards of professional excellence. This mechanism is not offset by greater transparency; quite the contrary, simply drawing attention to the use of AI triggers more critical scrutiny and greater skepticism.
Another striking aspect of this dynamic concerns the way in which the use of AI is revealed: when it is exposed by a third party—for example, through a leak or an external audit—the penalty in terms of trust is even greater than if the user had chosen to be transparent from the outset. This underlines the extent to which controlling one's communication around AI, and proactive or defensive image management, are becoming central to individual and collective strategies. We thus observe a gradation in the loss of trust: silence protects, disclosure weakens, and exposure ruins credibility.
This transparency paradox is accompanied by a form of social dissonance. Indeed, while the majority of individuals surveyed believe that the use of AI should be disclosed, they hesitate to apply it to themselves for fear of negative judgment. This social hypocrisy recalls the "privacy paradox" in digital behavior: we demand that others protect our data while adopting risky practices ourselves. Similarly, everyone is critical of others' use of AI, but remains discreet about their own use of these tools.
The in-depth analysis also reveals that the penalty for disclosing AI is little mitigated by familiarity with the technology: even those who are used to working with AI or who are regular users of it retain a certain distrust of people who explicitly announce their use of these solutions. However, there are two moderating factors: a very positive attitude towards technology in general and a perception of the reliability of AI slightly reduce the extent of the loss of trust, without however canceling it out entirely. It is therefore the deep representations of legitimacy and the role of humans that weigh the most in the balance.
The implications of these findings are numerous and concern managers, HR teams, communicators, and public decision-makers alike. Organizations are faced with a complex choice: should they make disclosure of AI use mandatory, at the risk of undermining trust in their employees, or should they leave it optional, at the risk of facing scandals in the event of non-consensual discovery? The answer is not simple and must take into account psychological reality: transparency is not always perceived as a virtue if the information revealed calls into question the foundations of legitimacy. Internal policies must therefore be accompanied by a deep reflection on the valorization of AI and the reconfiguration of social norms around augmented human work.
The phenomenon is also observed in the customer-brand relationship. When a company specifies in its marketing communications or processes that it uses AI, consumer trust suffers: the message is deemed less authentic, the brand less involved, and the product less unique. The economic benefits of AI then risk being offset, or even canceled out, by a deficit in trust and engagement. It then becomes essential to rethink how AI is integrated into the value proposition: should it be hidden, assumed, or educated so that the public perceives it as an asset? There is no indication today that simple technological progress will be enough to spontaneously change mentalities: the effort to support change is essential.
The case of AI autonomy sheds another light. When a task is performed exclusively by artificial intelligence, evaluators' confidence is higher than when a human discloses having used AI for support. This is explained by the clarity of roles: an autonomous agent, clearly identified as non-human, arouses a specific expectation, which does not cloud the perception of the legitimacy of human work. On the other hand, the unclear collaboration between human and machine introduces ambiguity: who is really the author of the result? This dilution of human responsibility and expertise fuels mistrust.
This study also raises exciting research prospects. As AI becomes more prevalent and its presence becomes commonplace, the question of how legitimacy standards are evolving becomes crucial. Younger generations, exposed to AI from school onwards, could see their relationship with authenticity and trust transform. It will also be interesting to observe whether, in highly automated fields, the penalty for disclosure is reduced or disappears, or whether new forms of "distrust" are recreated around other criteria. Researchers are encouraged to explore the diversity of contexts (health, finance, education, creative industry) in more detail and to integrate other psychological mechanisms, such as the perception of competence, integrity, or benevolence, into the formation of trust judgments.
Ultimately, the issue of AI disclosure is not simply a matter of administrative complexity or a uniform ethical imperative. It involves the very fabric of trust, which is based on the perceived legitimacy of practices and adherence to implicit norms. As AI continues to profoundly transform working methods, social expectations, and professional relationships, it is becoming urgent for organizations, leaders, and innovators to adopt a clear-eyed and nuanced approach to transparency. Integrating AI into practices is not enough: we must also reinvent the way we talk about AI, value augmented humans, and rebuild trust around a new alliance between artificial and human intelligence.
This challenge is all the more pressing because the dynamics of trust, once broken, are difficult to repair. Preserving the credibility of individuals and organizations, while benefiting from the real gains of AI, requires rethinking communication, internal culture, and talent management policies. The issue is therefore not simply technical or regulatory: it is profoundly human, at the intersection of social psychology, ethics, and business strategy.
To delve deeper: this paradox of trust echoes our article on generative AI and the GDPR and our analysis of Shadow AI.