Building trust in AI: From responsible governance to perceived fairness

On this page
Why trustworthy AI requires both accountable systems and fair decision-making experiences.
By Tina Shah Paikeday and Stephen W. Gilliland
Artificial intelligence is transforming how organizations make decisions about people, creating opportunities to improve efficiency, consistency, and fairness. Yet as AI becomes more deeply integrated into workforce decisions, leaders face a fundamental challenge: how can organizations ensure that AI systems are not only fair in practice, but also trusted by the people whose lives they affect?
This question sits at the intersection of responsible AI governance and organizational justice. While governance provides the structures, safeguards, and accountability necessary to manage AI responsibly, decades of research on organizational fairness demonstrate that trust depends on more than the technical quality of a decision. It also depends on how people perceive the process, the outcome, and the way they are treated.
Understanding both dimensions is essential for leaders seeking to build trust in an increasingly AI-enabled workplace.
Responsible AI governance is a leadership responsibility
AI presents significant opportunities to improve talent decisions, but those opportunities arrive alongside risks. AI can help organizations evaluate candidates more consistently, surface relevant evidence, and reduce reliance on subjective judgments. At the same time, poorly designed systems can amplify bias, obscure accountability, and introduce new concerns about privacy and transparency. Global research on public attitudes toward AI reinforces the importance of appropriate safeguards and governance in building trust (Gillespie et al., 2025).
Human decision-making is particularly susceptible to cognitive biases. Research on judgment and decision-making demonstrates how people rely on mental shortcuts that can produce systematic errors (Tversky & Kahneman, 1974). In hiring, intuitive judgments can be influenced by affinity bias, halo effects, and assumptions about credentials. Well-designed AI can support a shift toward more structured, criteria-based evaluation, helping decision-makers apply consistent standards at scale (Paikeday, 2025).
However, technology alone does not guarantee better decisions. The same systems that support fairness can perpetuate existing inequities when trained on biased information or implemented without appropriate safeguards. This makes responsible governance a leadership responsibility rather than simply a technical or compliance function.
At Findem, responsible AI governance is organized around three interconnected layers: governance and compliance, operational safeguards, and education. Grounded in the National Institute of Standards and Technology’s AI Risk Management Framework (NIST AI RMF), the approach addresses four core areas: employment fairness, AI transparency, data privacy, and information security (National Institute of Standards and Technology, 2023). Education creates a continuous feedback loop: insights from governance practices inform ongoing learning, while better-informed leaders and practitioners strengthen oversight by asking sharper questions, identifying risks, and applying responsible AI principles in practice.
An essential principle is that AI should recommend and explain while people retain responsibility for consequential decisions. Rather than functioning as a gate that automatically excludes candidates, AI can serve as a mirror, surfacing relevant evidence and making human judgments more transparent and open to examination. Research on persuasive technology and behavior change support systems reinforces the importance of designing technology with an understanding of how it influences human behavior (Fogg, 2003; Oinas-Kukkonen, 2013).
This approach also requires leaders to determine the appropriate level of AI autonomy. As systems evolve from assistive tools to agents capable of completing increasingly complex tasks, the human role shifts from directing individual actions to collaborating, supervising, or auditing. The appropriate balance depends on the nature of the decision and the risks involved (Raisch & Krakowski, 2021).
Trust requires independent verification
As AI capabilities advance, organizations are increasingly being asked not only what their systems can accomplish, but how those systems are governed. Trust depends on whether stakeholders believe appropriate safeguards exist and whether those safeguards can be independently verified (Gillespie et al., 2025).
The challenge is particularly visible among frontier AI developers, which have introduced safety frameworks intended to manage the risks associated with increasingly capable models. In September 2026, The Information reported that Anthropic, OpenAI, and Google had been discussing the creation of an industry standards body to support AI safety oversight (Schwartz, 2026). Yet an independent evaluation of twelve frontier AI safety frameworks found scores ranging from 8% to 35%, identifying gaps in areas such as measurable risk tolerances and clear thresholds for pausing development (Stelling et al., 2025). These developments highlight both the growing recognition of the importance of AI governance and the limitations of relying exclusively on industry self-regulation.
For organizations deploying AI in workforce decisions, independent verification is equally important. At Findem, third-party audits are used to evaluate employment fairness and information security. Warden AI audits AI-assisted scoring for disparate impact, while SOC 2 Type II assessments provide independent assurance around security controls.
However, auditing an algorithm is only part of evaluating fairness. The broader decision-making process must also be examined because human behavior can undermine even a technically sound system.
Research demonstrates that evaluators may redefine selection criteria to favor candidates they already prefer, a phenomenon known as criteria drift (Uhlmann & Cohen, 2005). The design of an interface can also influence decisions through rankings, defaults, and the presentation of information (Thaler & Sunstein, 2008). Meanwhile, collecting independent judgments before group discussion can reduce noise and the influence of others’ opinions (Kahneman et al., 2021).
Responsible AI governance must therefore examine the combined human-AI decision-making system, not simply the model operating within it.
Fairness is both a reality and a perception
Even when organizations establish rigorous governance practices, a critical question remains: will people perceive the resulting decisions as fair?
Research on organizational justice demonstrates that perceptions play a more significant role in how people understand and evaluate fairness than we might assume. Individuals evaluate fairness based on their own experiences, expectations, and comparisons with others.
The consequences extend well beyond an individual’s immediate reaction. Perceived fairness can have a ripple effect, increasing acceptance of decisions, trust in leadership, organizational commitment, and engagement. Conversely, perceived unfairness can undermine confidence in leaders and reduce willingness to accept organizational decisions.
Organizational justice research identifies three interconnected dimensions of fairness:
Outcome fairness concerns what people receive relative to their contributions and expectations. In employment decisions, individuals evaluate whether outcomes such as hiring, promotions, or compensation are appropriate, often comparing their experiences with those of others.
Process fairness concerns how decisions are made. Fair processes use accurate and relevant information, apply criteria consistently, minimize bias, and provide opportunities for individuals to offer input or appeal decisions. These principles build on Leventhal’s (1980) procedural justice framework and subsequent research on fairness in employee selection (Gilliland, 1993).
Interactional fairness concerns how individuals are treated throughout the decision-making process. Respect, dignity, transparency, explanations, and appropriate acknowledgment of unfavorable outcomes all shape whether people experience a decision as fair.
Importantly, fair processes and respectful treatment can improve acceptance even when the outcome is unfavorable. For organizations using AI in hiring and other workforce decisions, this means that fairness cannot be evaluated solely through the accuracy or statistical performance of an algorithm. The experience of the person receiving the decision matters as well.
Why people may perceive AI as fairer than humans
Emerging research raises a particularly interesting question: do people sometimes perceive AI decisions as fairer than decisions made by human managers?
A series of studies by Choi and Chao (2026) examined reactions to favorable and unfavorable decisions made by AI versus human decision-makers. Their research found that AI decisions could generate stronger perceptions of fairness, contributing to greater decision acceptance and trust.
The researchers examined whether AI’s perceived fairness advantage could be explained by perceptions of greater objectivity. However, this relationship was not statistically significant. Instead, the findings indicated that AI was perceived as less emotional than human managers, and this perception contributed to greater judgments of fairness (Choi & Chao, 2026).
This distinction matters. People may associate human decision-makers with emotions, personal preferences, and inconsistent judgments, while viewing AI as less susceptible to these influences. As a result, AI can benefit from a perception of fairness even when individuals do not have direct evidence that its decisions are objectively more equitable.
The findings also reveal how fragile these perceptions can be. When participants were reminded that AI systems could be trained on biased data, the perceived fairness advantage diminished. Furthermore, the research represents one published set of studies, and attitudes toward AI are changing rapidly. Additional research is necessary to determine whether these findings will persist as public familiarity with AI increases.
For leaders, the implication is significant: perceived fairness should not be mistaken for demonstrated fairness. An AI system may inspire confidence because it appears unemotional or impartial, but that confidence must be supported by evidence of actual fairness, transparency, and appropriate oversight.
The human-AI partnership can introduce new fairness challenges
Retaining human judgment in AI-assisted decisions is often presented as an important safeguard. Yet human involvement does not automatically produce fairer outcomes.
Consider a selection system that has been scientifically validated and shown to be free from adverse impact. If an employer independently screens resumes using subjective judgment before candidates enter the system, they risk reintroducing the very biases the system was designed to prevent, undermining the integrity of an otherwise fair selection process.
This illustrates a central challenge of human-AI collaboration: organizations must determine where technology can improve consistency and where human judgment adds meaningful value.
AI may be particularly effective at processing large volumes of information against established criteria, a task that humans can struggle to perform consistently because of fatigue, time constraints, or cognitive biases. Human expertise, meanwhile, remains important for contextual judgments that require a deeper understanding of organizational needs, interpersonal dynamics, and complex circumstances.
The goal is not to eliminate human judgment, but to apply it deliberately. Leaders must establish clear boundaries for when people should rely on AI-generated evidence, when they should exercise independent judgment, and how deviations from established criteria should be evaluated (Raisch & Krakowski, 2021).
This distinction becomes especially important when assigning accountability. People may hold human decision-makers responsible for unfavorable outcomes while attributing successful decisions to the perceived objectivity of AI. Although this possibility requires further empirical study, it highlights the importance of clearly defining responsibility within human-AI decision processes.
Building trust through governance, education, and fair treatment
Responsible AI leadership requires organizations to connect technical governance with the human experience of fairness. Independent audits, transparent explanations, consistent decision criteria, and opportunities for review can help establish objectively defensible processes. Equally important are the opportunities for individuals to understand decisions, express concerns, and receive respectful treatment (Gilliland, 1993; Leventhal, 1980).
Education plays an essential role in connecting these dimensions. Leaders, employees, and users must understand not only how AI works, but also when to trust its outputs and when to question them. The appropriate level of trust depends on the role AI plays, the evidence supporting its performance, and the consequences of relying on it (Fogg, 2003; Gillespie et al., 2025).
Ultimately, the objective is not to maximize trust in AI. It is to ensure that trust is appropriately calibrated to the system’s actual capabilities and safeguards.
As AI becomes a more influential participant in workforce decisions, organizations have an opportunity to strengthen fairness by combining the consistency of technology with the accountability and contextual understanding of people. Achieving that balance requires more than deploying capable systems. It requires governance that can be verified, decision processes that can be explained, and an ongoing commitment to the people affected by those decisions.
Trustworthy AI is not simply AI that people believe is fair. It is AI whose fairness can be demonstrated, whose decisions can be questioned, and whose use reinforces trust in the organizations responsible for it.
References
Choi, J., & Chao, M. M. (2026). For me or against me? Reactions to AI (vs. human) decisions that are favorable or unfavorable to the self and the role of fairness perception. Personality and Social Psychology Bulletin, 52(3), 671–691.
Fogg, B. J. (2003). Persuasive technology: Using computers to change what we think and do. Morgan Kaufmann.
Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. https://doi.org/10.26188/28822919
Gilliland, S. W. (1993). The perceived fairness of selection systems: An organizational justice perspective. Academy of Management Review, 18(4), 694–734. https://doi.org/10.2307/258595
Kahneman, D., Sibony, O., & Sunstein, C. R. (2021). Noise: A flaw in human judgment. Little, Brown Spark.
Leventhal, G. S. (1980). What should be done with equity theory? New approaches to the study of fairness in social relationships. In K. J. Gergen, M. S. Greenberg, & R. H. Willis (Eds.), Social exchange: Advances in theory and research (pp. 27–55). Plenum Press.
National Institute of Standards and Technology. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). U.S. Department of Commerce. https://doi.org/10.6028/NIST.AI.100-1
Oinas-Kukkonen, H. (2013). A foundation for the study of behavior change support systems. Personal and Ubiquitous Computing, 17(6), 1223–1235. https://doi.org/10.1007/s00779-012-0591-5
Paikeday, T. S. (2025). Slow thinking fast: How AI trumped human bias. California Management Review. https://cmr.berkeley.edu/2025/06/slow-thinking-fast-how-ai-trumped-human-bias/
Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation–augmentation paradox. Academy of Management Review, 46(1), 192–210. https://doi.org/10.5465/amr.2018.0072
Schwartz, L. (2026, September). Anthropic, OpenAI, Google quietly discussed an AI safety standards body. The Information. https://www.theinformation.com/articles/inside-ai-industrys-behind-scenes-push-police
Stelling, L., Murray, M., Campos, S., & Papadatos, H. (2025). Evaluating AI companies’ frontier safety frameworks: Methodology and results [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2512.01166
Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. https://doi.org/10.1126/science.185.4157.1124
Uhlmann, E. L., & Cohen, G. L. (2005). Constructed criteria: Redefining merit to justify discrimination. Psychological Science, 16(6), 474–480. https://doi.org/10.1111/j.0956-7976.2005.01559.x












