With the rapid advancement of artificial intelligence, the concept of trustworthy AI (TAI) has gained significant prominence, and various guidelines have emerged to direct the development of trustworthy systems. While research has produced valuable insights about TAI, we lack a comprehensive understanding of its characteristics and for whom the current TAI frameworks are important. In this paper, we address this challenge through a scoping review of the concept of TAI, proposing distinct characteristics of TAI based on the their themes and identifying the stakeholders for whom current frameworks are most relevant. This paper contributes to the literature on AI-systems development and deployment by ofering a comprehensive understanding of trustworthy AI. In addition, it highlights the challenges of translating the concept of trustworthy AI into practice and what this means across diferent stakeholder groups.
Artificial Intelligence (AI) has rapidly evolved into a transformative research field, attracting
significant attention from both academia and industry. This growing interest is reflected in
the expansion of the AI market, driven by advancements in machine learning, increasing
availability of data, and a rising demand for intelligent systems across various industries [
In recent years, numerous frameworks, guidelines, and reports have emerged to establish
trust in AI. Various organizations worldwide have developed ethical guidelines and principles
to establish trust in AI systems [
Although numerous guidelines exist, the conceptualization of TAI remains inconsistent across disciplines, and terminological variations further contribute to fragmentation. Even within the same terminology, diferent sources ofer slightly diferent interpretations of TAI principles, requirements, and recommendations. This inconsistency fosters the perception that AI communities lack a unified understanding of TAI, complicating eforts to build trust across stakeholders. However, despite these challenges, these guidelines and principles remain crucial in fostering trust throughout the life-cycle of the AI system.
Given the increasing interest in AI’s societal impact and its ethical alignment, TAI has emerged as a key concept in addressing critical concerns about AI governance. To bridge the existing research gap, this literature review synthesizes the current state of knowledge on TAI in information systems and related fields such as computer science. As the discourse on TAI continues to expand, it is imperative to consolidate existing research to provide a strong foundation for both scholars and practitioners. This synthesis aims to clarify key concepts, identify knowledge gaps and challenges, and highlight opportunities for future research. In this context, we formulate the following research question:
By addressing this question, the study contributes to a more comprehensive understanding of TAI, and highlights the importance of considering whose trust is impacted by AI systems, emphasizing the diverse stakeholders and contexts in which trust operates.
Trust is relational and complex, involving at least two actors: one who trusts (the trustor) and one who is trusted (the trustee). In this relationship, the trustor relies on the trustee to carry out (or not carry out) a particular action [16]. Numerous definitions of trust exist across diferent disciplines, but Lukyanenko et al. [17] ofer a general definition by generalising human mental and physiological mechanisms to trust agents, such as AI agents. They define trust as “general trust is an information processing and behavioural process within a trusting agent that considers the properties of another system to control the extent and parameters of the interaction with this system.” In the context of Information Systems, Mayer et al. [18] describe trust as “trust is the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party.” Duenser and Douglas [19] add that “trust relationships involving AI are socio-technical in nature, incorporating not only the AI itself but also people, laws, social norms, and institutions.” Similarly, Siau and Wang [20] propose three perspectives on trust: (1) a set of specific beliefs dealing with benevolence, competence, integrity, and predictability (trusting beliefs); (2) the willingness of one party to depend on another in a risky situation (trusting intention); or (3) the combination of these elements.” While closely related, trust and trustworthiness are distinct concepts. Trustworthiness, on the other hand, is a trait (characteristic), often confused with trust itself. Being trustworthy does not automatically create a trust relationship, nor does it guarantee that trust will be established [16]. The following section elaborates on the transition from trustworthy computing to TAI.
In the United States, 1999 report “Trust in Cyberspace” by the National Academies [21, 22] laid the foundations for what is now called “trustworthy computing” as a key area of research. Shortly after, the national science foundation (NSF) launched a series of initiatives to advance this field, starting with the “trusted computing program” in 2001, followed by the “cyber trust initiative” in 2004. By 2007, the trustworthy computing program was introduced, and in 2011, it evolved into the secure and trustworthy cyberspace initiative (SaTC). Spearheaded by the NSF’s Computer and Information Science and Engineering Directorate, these eforts expanded trustworthy computing research beyond computer science into an interdisciplinary endeavor.
Industry has also played a central role in shaping the trustworthy computing landscape, with industry leaders like Microsoft at the forefront. In January 2002, Bill Gates issued his famous “Trustworthy Computing” memo [23], which marked a turning point for Microsoft and the broader tech industry. Gates’ directive highlighted the urgent need for the development of trustworthy software and hardware products. The memo drew upon an internal Microsoft white paper that defined four key pillars of trustworthiness: security, privacy, reliability, and business integrity. This shift not only influenced Microsoft’s approach but also set a precedent for the broader IT sector, solidifying the importance of trust as a core element of modern computing practices. According to Wing [22], trustworthy computing comprises a set of overlapping properties—reliability, safety, security, privacy, availability, and usability—applicable to hardware and software systems and their interactions with humans and the physical world. Wing [22] also emphasized that TAI systems require a more comprehensive set of properties than traditional computing systems. TAI encompasses not only reliability, security, privacy, and usability, but also additional properties such as probabilistic accuracy under uncertain conditions, fairness, robustness, accountability, and explainability.
The term “artificial intelligence” was first conceived at a workshop at Dartmouth college in
1956 [24]. Since then, the field has undergone several waves of rapid progress [ 25]. Since the
early 2010s, machine learning and deep learning have achieved groundbreaking advancements,
with the pace of progress continuing to accelerate. These developments have fueled visions
of a world enriched by intelligent agents enhancing individual’s lives, organizations, and
societies. However, AI is not a universal solution or a magic bullet. Like any other technology,
it ofers significant benefits while also introducing new ethical, legal, and social challenges
that must be carefully addressed [26, 27]. The growing recognition of these challenges has led
to the proliferation of AI frameworks, guidelines, and reports in recent years. Notably, two
well-established public repositories of AI ethics guidelines include algorithmwatch [28] and
linking AI principles (LAIP) [29]: Algorithmwatch, an organization that monitors the societal
impact of digitalization, curated a collection of 167 AI ethics guidelines before ceasing new
submissions in April 2024 [28]. Meanwhile, LAIP continues to compile ethical frameworks, with
115 proposals recorded as of April 2, 2025 [29]. Overall, these guidelines and recent research on
trustworthiness [30] in the context of AI provide the key foundation for exploring the landscape
of TAI in this review. These frameworks and guidelines aim to guide the design, development,
and implementation of AI systems in ways that benefit individuals, businesses, and society while
reinforcing human-centric values. However, diferent concepts and terminologies are often
interpreted diferently by various users and organizations. AI systems encompass both technical
artefacts and human operators [31, 32] and not isolated; they are embedded within social
contexts, forming complex socio-technical systems where society, technology, and organisation
mutually shape and influence one another [33]. There are various definitions of TAI, however
according to the European Commission’s AI high-level expert group (HLEG), TAI must comply
with legal, ethical, and technical standards while also being socially robust [
Trust is a fundamental factor in the moral and ethical treatment of stakeholders. Hosmer describes trust as “the expectation by one person, group or firm of ethically justifiable behavior on the part of the other person, group or firm in a joint endeavor or economic exchange [ 34].” In the context of TAI systems, a diverse range of stakeholders, including designers, developers, AI/ML experts, data scientists, system engineers, project and product managers, regulators, funding organizations, auditors, and users of AI technology, among others, all play a role [35].
The responsibility for ensuring the development and use of TAI systems extends to individuals and organizations involved in their design, development, and maintenance, as well as legislative and regulatory bodies at both the national and international levels [35]. From a practical point of view, stakeholders directly influencing TAI system development include those responsible for designing, building, and maintaining the systems, as well as the organizations funding their creation. In contrast, those shaping ethical AI guidelines, such as public, private, and non-profit organizations, and those involved in developing legal and regulatory frameworks for AI, play an indirect role in shaping the behavior of AI systems. These actors must collaborate to establish rules and frameworks that address the impacts of TAI. However, as the ethical AI guidelines gain further clarity, the influence of these stakeholders, across public, private, and non-profit sectors, will become more pronounced. Similarly, as national and international regulatory bodies develop more detailed legal obligations for TAI systems, their impact on the day-to-day practices of TAI development and deployment will continue to grow.
Despite the interest in developing TAI, there have been insuficient eforts to synthesize existing research on how it is thought about, as well as to evaluate how current frameworks address the needs and perspectives of various stakeholders (trustors) [36]. By broadening the scope to include diverse stakeholder perspectives, we can better address relational aspects of trust and ensure that TAI is designed to meet the needs of those it impacts.
This literature review was conducted to explore current literature on trustworthy AI, its elements and relevance to various stakeholders. The chosen review methodology is inspired by the approach outlined by Webster & Watson [37], which ofers valuable guidance for conducting a concept-centric analysis that synthesizing existing knowledge.
• Papers focused on purely computational or technological work. • Papers addressing only specific elements of TAI rather than the whole concept or other terminologies except those that seem to be very close.
We began the search process by selecting keywords, which we then used for literature searches. This was followed by evaluating the identified literature and selecting relevant articles. Finally, we conducted an in-depth analysis of the chosen literature. 3.2.1. Keyword Selection We began by conducting a ‘traditional’ search with a broad query to identify fundamental keywords for discussing TAI’s elements or dimensions, along with the relevant stakeholders involved. The reason for the broad query was to avoid limiting the number of hits and to identify the other concepts related to trustworthy AI. The search string, executed in the Scopus and Web of Science (WoS) databases (‘trustworthy artificial intelligence’ OR TAI OR ethics) AND (stakeholder*), helped identify articles addressing the ethical dimensions of TAI, including Inclusion Criteria: Exclusion Criteria: • Research papers that discuss or theorize concepts related to TAI (interdisciplinary field). • Papers on TAI and related ethical guidelines or principles. synonyms of TAI, and the stakeholders involved to achieve TAI. The asterisk was used to ensure that diferent variations of word conjugations were included in the search (e.g., ‘stakeholder,’ and ‘stakeholders’). Additionally, the ‘literature review papers’ discovered during the keyword search proved useful for finding related keywords. 3.2.2. Literature Search We expanded the search query to include more than just trustworthiness-related terms, as certain terms can serve as synonyms for trustworthy AI. Additionally, we incorporated stakeholder terms to meet the requirements of this research work. Consequently, the query included:
Query: (“trustworthy artificial intelligence” OR TAI OR “responsible artificial intelligence” OR RAI OR “explainable artificial intelligence” OR XAI OR “ethical artificial intelligence” OR “trustworthy AI” OR “responsible AI” OR “explainable AI” OR “ethical AI” OR “lawful AI” OR “lawful act” OR “ethical guideline*” OR “ethical principle*” OR “ethical standard*”) AND (stakeholder*).
Various combinations of keywords were employed to ensure relevancy in the search. The publications were collected from the digital repositories of the association for information systems AIS eLibrary (primarily to cover IS conference papers), Scopus database (using “https://litbaskets.io/” to identify relevant IS and related interdisciplinary journals), Web of Science (WoS), and the ACM Digital Library (Computing Literature).
The searches were performed on the title, abstract, and keywords of papers using the Boolean operators “OR” and “AND.” The AIS eLibrary was searched due to its comprehensive coverage of the latest advancements in both practice and academia within IS research. Since IS is an interdisciplinary field that straddles other disciplines, it is often necessary to look not only within the IS discipline when reviewing and developing theory but outside the field (Webster & Watson, 2002). Therefore, the WoS, Scopus, and the ACM were also searched, which provide access to a diverse array of peer-reviewed literature, including academic journals, conference proceedings, and other scholarly documents across various academic fields.
An initial set of 896 records was compiled from four databases: 161 from the ACM Digital Library, 716 from Web of Science, 19 from Scopus (via Literature Baskets), and none from AIS eLibrary. 3.2.3. Literature Evaluation and Selection The literature search was conducted with the aim to ensure comprehensive coverage of the existing literature. The articles are restricted to English language and only included peerreviewed publications to ensure the research quality. The collection period for the AIS eLibrary, ACM, and WoS was from March 1, 2020, to March 31, 2025. For Scopus, due to date limitations, publications were collected from 2020 to 2025. The following items were excluded: articles that did not meet the publication date criteria, papers outside the search scope, books or book sections, articles shorter than six pages. Additionally, duplicates and triplicates were carefully avoided, and references were manually examined to uncover any additional articles or papers. The selection of papers was conducted in three steps. First, all fetched articles were screened to assess their eligibility based on the inclusion and exclusion criteria. In the second step, the abstracts of the remaining publications were reviewed. In the third step, the full text of all remaining papers was thoroughly examined to identify those to be included. Finally, backward and forward searches were conducted using the papers identified as starting points, yielding two more relevant studies—one from each method. Throughout these steps, the inclusion and exclusion criteria were consistently applied. In total, twenty papers from the primary, backward, and forward sets were selected following in-depth review and filtering [35, 36, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55]. We used the PRISMA lfow diagram [56] for clear and transparent reporting of the search process (see Figure 1).
In addition, sixteen frameworks, principles, or guidelines cited in the selected articles were analyzed, ofering valuable insights into the components of TAI and related concepts. While some of these sources were published before 2020, it is important to emphasize that the primary goal of this research is not to provide an exhaustive review of all global AI policy frameworks, principles, or ethical guidelines. Rather, it aims to identify the key elements that define TAI and who are the stakeholders.
Table 1 categorizes the articles by terminology and publication year. Among the terms used, Explainable AI (XAI) is the most cited term (six articles), followed by Responsible AI (RAI) with four. Trustworthy AI (TAI), Ethical AI (EAI), and Human-Centric AI (HC-AI) appear in two articles each. Most publications appeared in 2024 (eight articles), followed by 2023 (five) and 2021 (four). One article was published in each of 2025, 2022, and 2020, totaling twenty articles.
These TAI frameworks include AI HLEG (4 principles), OECD (5 principles), White House OSTP (10 principles), G20 AI principles (5 principles). RAI: Chinese AI Principles (8 principles), Facebook (5 principles), and Montreal Declaration (10 principles). EAI includes UNESCO’s Recommendation (10 principles), IBM’s Everyday EAI (5 principles), AI4People (5 principles), and UK AI Code (5 principles). HC-AI is represented by Japan’s social principles (7 principles), XAI includes NIST’s XAI framework (4 principles), while beneficial AI (BAI) includes Asilomar AI Principles (23 principles). These frameworks were further validated using two well-established public repositories of AI ethics guidelines: AlgorithmWatch [28] and the LAIP repository [29]. Overview of the core TAI and related terminologies and associated elements
Table 3 presents a comprehensive overview of the ten selected AI frameworks, principles, or ethical guidelines, along with their associated elements. In the final column of Table 3, the frequency of each element mentioned across diferent frameworks having various themes. The last row of Table 3 displays the ratio of partially included ethical elements1 relative to the total number of principles/requirements outlined in the original policy frameworks.
In this section, we discuss the results, including key elements of TAI, the stakeholders involved in achieving TAI, and identify areas that have not been suficiently explored. It also proposes the future directions for researchers and practitioners to advance the development of TAI. The discussion covers AI-community fragmentation, TAI and its key elements, stakeholders involved, and the limitations of the current approach.
AI, like any technology, ofers significant benefits but also introduces new ethical, legal, and social challenges [26]. In response, numerous guidelines, frameworks, or guiding principles related to TAI and its related concepts have emerged in recent years. These frameworks often propose varying numbers of principles, requirements, or recommendations. Notably, diferences arise not only between distinct AI-related terminologies but also within the same terminology as defined by diferent sources (see Table 2). As a result, the landscape of AI governance has become increasingly complex, with various overlapping and sometimes inconsistent attributes expected from AI systems. Terms like ‘AI Safety,’ ‘Fairness in AI,’ ‘Secure AI,’ ‘Explainable AI,’ ‘Transparent AI,’ ‘Responsible AI,’ ‘Trustworthy AI,’ ‘Interpretable AI,’ ‘Robust AI,’ ‘Ethical AI,’ ‘Accountable AI,’ ‘Resilient AI,’ ‘Reliable AI,’ ‘Black-box AI,’ ‘Privacy-enhanced AI,’ and ‘Federated AI’ reflect this diversity of perspectives in this space [57]. This diversity can complicate understanding and create the impression that AI governance communities are fragmented, and that the concept of TAI lacks coherence and a unified definition. Nevertheless, despite these terminological diferences, these initiatives share a common objective: to enhance the benefits of AI while mitigating its associated risks and harms. Ultimately, these overlapping concepts reflect both multidisciplinary and interdisciplinary eforts to guide the development and deployment of TAI systems, guided by experts from various disciplines. Although many definitions of TAI have been proposed across various scenarios and themes, there is still no universally accepted definition of TAI. The following are some of the most commonly used TAI-related terms and their definitions, developed by experts from diferent disciplines and in diferent contexts.
Trustworthy AI: As defined by the european commission’s high-level expert group (HLEG)
on AI, trustworthy AI must adhere to legal, ethical, and technical standards while also being
socially robust [
Responsible AI: According to the world economic forum (WEF), responsible AI refers to the practice of designing, building, and deploying AI in a manner that empowers people 1Here partially included elements refer to cases where certain elements within specific frameworks serve as subelements under diferent themes, often interpreted with slight variations. and businesses while ensuring fair impacts on customers and society [58]. This approach enables companies to engender trust and scale AI with confidence [ 58]. Arrieta et al. further define responsible AI as a methodology for the implementation of AI methods in real world organization with fairness, model explainability and accountability at its core [59].
Explainable AI: As noted by defense advanced research projects agency (DARPA), explainable AI aims to produce more explainable models, that maintain high learning performance (prediction accuracy) while enabling human users to understand, appropriately trust, and efectively manage AI systems [15].
Ethical AI, as defined by the european commission’s high-level expert group (HLEG) on AI,
ethical AI refers to the development, deployment and use of AI that ensures compliance with
ethical norms, including fundamental rights as special moral entitlements, moral principles and
related core values [
Despite difering terminologies, these initiatives share the common goal of maximizing AI’s benefits while mitigating risks and potential harm.
The goal of this review is to enhance understanding of TAI, identify its key elements tailored to various contexts or themes, and identify the associated stakeholders. To explore the key elements of TAI, we began by reviewing ten AI policy frameworks, guiding principles, or guidelines (see Table 3), tailored to diferent thematic areas (some listed in Table 2) and introduced by various global organizations. These documents define the essential elements that AI systems must possess to be considered trustworthy. Categorizing these socio-technical elements proved challenging, as elements identified as primary in one framework often appeared as sub-elements in another. Moreover, these elements were often organized under diferent context or themes, sometimes with slight variations in interpretation. Considerable efort was invested in organizing these elements and assessing their frequency of occurrence across the frameworks. Among the most frequently cited elements—appearing in at least eight of the ten frameworks—were:
Accountability: A relationship between an actor and a forum, in which the actor has an
obligation to explain and to justify his or her conduct, the forum can pose questions and pass
judgement, and the actor may face consequences [60]. Transparency: It refers to the need
to explain, interpret, and reproduce its decisions [61]. Privacy and data governance: It
emphasizes protecting data privacy, integrity, and quality, while ensuring individual’s rights
to access their data. [62]. Technical robustness and safety: This means, AI systems should
be technically robust and perform as expected by the users [
However, Jobin et al. [
TAI elements are often interrelated; for instance, transparency supports accountability, and fairness contributes to societal well-being. However, balancing these values in practice frequently requires design trade-ofs. A common tension arises between explainability and performance, where more interpretable models—such as decision trees—may underperform compared to more complex, deep neural network models. Similarly, eforts to enhance fairness, such as mitigating bias, can sometimes lead to reduced accuracy for the majority class. The tension between privacy and utility is also well-documented; for example, implementing differential privacy may protect individual data but reduce the utility of datasets. Overall, these ifndings highlight the core elements commonly emphasized in TAI discourse, suggesting that existing frameworks may structure these elements diferently depending on the thematic focus they are developed around and the stakeholders involved. This observation can be seen as a contribution to the field, ofering a batter understanding of the evolving landscape of TAI.
Duenser and Douglas [19] stated that: “Trust relationships involving AI are socio-technical in nature, incorporating not only the AI itself but also people, laws, social norms, and institutions.” Viewing AI as part of a socio-technical system is essential, as the roles of people (the stakeholders) involved, those who design, develop, deploy, governance, guide and use the technology, are as critical as the technology itself in establishing its trustworthiness.
Viewing TAI through a stakeholder lens enhances our socioeconomic understanding of the primary challenges in achieving TAI. It emphasizes the interdependency among various stakeholders, the themes for which TAI systems are developed, the technology itself, and the social contexts in which TAI systems are developed and deployed. This perspective shifts the focus from seeing AI merely as a technical artifact to recognizing TAI as a system deeply embedded within human, organizational, and societal structures. Consequently, AI technologies must be understood within their socio-technical contexts, with a particular emphasis on stakeholders, rather than shareholders — both during their development and application.
From a practical perspective, the primary stakeholders directly influencing the development of TAI systems include those involved in designing, developing, and maintaining these systems, as well as the organizations that fund their creation. Those shaping the ethical AI guidelines, such as public, private, and non-profit organizations, and those involved in developing legal and regulatory frameworks for AI, play an indirect role in shaping the behavior of AI systems. Additionally, the ecosystem of TAI stakeholders extends to include not only the users of these systems but also those afected by their deployment. TAI, therefore, functions as a socio-technical system, encompasses diferent stakeholders including developers, its users, the technology, and the institutions that govern the interactions among diferent stakeholders. The term “institutions that shape these interactions” refers to the formal and informal organizations, policies, regulations, cultural norms, and frameworks that shape how TAI technologies are developed, deployed, and used. These can include governmental bodies, regulatory agencies, industry standards organizations, educational institutions, and even societal norms or ethical guidelines. These institutions establish rules (such as EU AI act), provide oversight, and set expectations that influence how stakeholders engage with AI systems. Efective collaboration among these actors is crucial for building sustainable TAI systems and for continuous updating and implementing rules and frameworks that address the broader impacts of TAI.
It is evident that various AI policy frameworks, principles, or guidelines have been developed with varying thematic considerations in mind, and none can be considered perfect or complete for every scenario. While, these documents present high-level objectives for TAI systems and the underlying science and technology, they do not delve into specific technical implementations. Moreover, the frameworks show significant overlap, with widespread agreement that AI should promote the common good, avoid causing harm or violating rights, and uphold fundamental human values.
Diferent stakeholders often have conflicting priorities. Such as, developers often prioritize optimizing model performance, whereas regulators and users emphasize concerns about systemic biases. Organizations may resist transparency due to proprietary algorithms, while civil society groups and users increasingly demand explainability and accountability. In domains such as health or mobility, researcher’s desire for open data access may conflict with individual’s rights to privacy and consent. Moreover, the application of international AI principles may clash with local legal systems, cultural values, or societal norms, creating friction across geopolitical and social contexts. To address these conflicts, we believe all stakeholders must come together to engage in dialogue, negotiate trade-ofs, and collaboratively develop solutions.
As ethical AI guidelines continue to evolve and become more defined, the influence of stakeholders across the public, private, and non-profit sectors is expected to grow significantly. Similarly, national and international regulatory bodies will play an increasingly important role as they introduce more detailed legal obligations with stakeholder responsibilities, further shaping the daily practices involved in the development and deployment of TAI systems.
This study limited to four databases—AIS eLibrary, scopus (via litbaskets), web of science, and ACM digital Library— may not have been exhaustive, but it is expected that these studies will contribute suficient knowledge, aiding both novice and experienced researchers within the studied subject. Furthermore, this study does not cover prior research on TAI-related topics published before March 1, 2020, except for those in scopus databases, due to specific date constraints (covering 2020–2025).
This study was motivated by the growing importance of TAI and the lack of a comprehensive understanding of its key elements and the stakeholders involved. In response, we conducted a scoping review to examine how TAI is conceptualized, what core TAI elements are emphasized, and which stakeholders are involved.
Through thematic analysis, we identified twenty-one distinct elements. Among the most frequently cited elements—appearing in at least eight of the ten frameworks—were accountability, privacy and data governance, transparency, human-centered values, technical robustness and safety, and diversity, non-discrimination, fairness, and democracy. Other important elements—including societal and environmental well-being, inclusive growth and sustainability, fairness and justice, explainability, and security—were cited in at least six frameworks. Additionally, human agency and oversight appeared in five frameworks. These elements are drawn from ethical frameworks developed around diverse themes, reflecting diferent contextual priorities and stakeholder perspectives. The diversity in structure and slight variations in interpretation reveal the challenges of consistently categorizing elements and highlight the potential ambiguities in correctly identifying elements and their corresponding stakeholder relevance.
Our findings underscore that TAI is not merely a technical challenge but a socio-technical endeavor involving a diverse ecosystem of stakeholders. These include developers, designers, funding organizations, policymakers, regulatory agencies, users, and afected communities. The study shows that ethical principles are often framed diferently depending on institutional priorities, cultural norms, and sectoral interests—resulting in distinct thematic orientations across frameworks. Consequently, understanding TAI elements and stakeholders within their specific contexts is essential for efectively interpreting and applying these frameworks. Viewing AI through a stakeholder lens enhances our socioeconomic understanding of the key challenges in achieving TAI. It highlights the inter-dependencies among stakeholders, the purposes for which AI systems are developed, the technologies themselves, and the social contexts in which they are implemented.
This observation can be seen as a contribution to the field, ofering a batter understanding of the evolving landscape of TAI, and the stockholders involved achieving TAI. We encourage further exploration, validation of these findings, and the incorporation of new insights to advance the ongoing discourse on TAI.
This research was supported by the Wallenberg AI, Autonomous Systems and Software Program – Humanity and Society (WASP-HS), funded by the Wallenberg Foundations, Sweden.
During the preparation of this work, the author(s) used ChatGPT-4 (OpenAI) to check grammar and spelling. The author(s) subsequently reviewed and edited the content as needed and take(s) full responsibility for the publication’s content. artificial intelligence–overview of trustworthiness in ai, https://www.iso.org/standard/ 77608.html, 2020. Standard. [15] D. Gunning, D. Aha, Darpa’s explainable artificial intelligence (xai) program, AI magazine 40 (2019) 44–58. [16] F. Gille, A. Jobin, M. Ienca, What we talk about when we talk about trust: theory of trust for ai in healthcare, Intelligence-Based Medicine 1 (2020) 100001. [17] R. Lukyanenko, W. Maass, V. C. Storey, Trust in artificial intelligence: From a foundational trust framework to emerging research opportunities, Electronic Markets 32 (2022) 1993– 2020. [18] R. C. Mayer, J. H. Davis, F. D. Schoorman, An integrative model of organizational trust,
Academy of management review 20 (1995) 709–734. [19] A. Duenser, D. M. Douglas, Whom to trust, how and why: untangling artificial intelligence ethics principles, trustworthiness, and trust, IEEE Intelligent Systems 38 (2023) 19–26. [20] K. Siau, W. Wang, Building trust in artificial intelligence, machine learning, and robotics,
Cutter business technology journal 31 (2018) 47. [21] N. R. Council, et al., Trust in cyberspace, National Academies Press, 1999. [22] J. M. Wing, Trustworthy ai, Communications of the ACM 64 (2021) 64–71. [23] B. Gates, Trustworthy computing, Wired News, Jan 17 (2002). [24] J. McCarthy, M. L. Minsky, N. Rochester, C. E. Shannon, A proposal for the dartmouth summer research project on artificial intelligence, august 31, 1955, AI magazine 27 (2006) 12–12. [25] M. Haenlein, A. Kaplan, A brief history of artificial intelligence: On the past, present, and future of artificial intelligence, California management review 61 (2019) 5–14. [26] L. Floridi, J. Cowls, M. Beltrametti, R. Chatila, P. Chazerand, V. Dignum, C. Luetge, R. Madelin, U. Pagallo, F. Rossi, et al., Ai4people—an ethical framework for a good ai society: opportunities, risks, principles, and recommendations, Minds and machines 28 (2018) 689–707. [27] L. Floridi, Establishing the rules for building trustworthy ai, Ethics, Governance, and
Policies in Artificial Intelligence (2021) 41–45. [28] A. Watch, The ai ethics guidelines global inventory, 2019. URL: https://inventory.
algorithmwatch.org/, accessed: 2025-03-18. [29] Laip - linking artificial intelligence principles, https://www.linking-ai-principles.org/, 2018.
Accessed: 2025-03-18. [30] M. M. Ferdaus, M. Abdelguerfi, E. Ioup, K. N. Niles, K. Pathak, S. Sloan, Towards trustworthy ai: A review of ethical and robust large language models, arXiv preprint arXiv:2407.13934 (2024). [31] I. Van de Poel, Embedding values in artificial intelligence (ai) systems, Minds and machines 30 (2020) 385–409. [32] P. Di Maio, Towards a metamodel to support the joint optimization of socio technical systems, Systems 2 (2014) 273–296. [33] E. L. Trist, The evolution of socio-technical systems, volume 2, Ontario Quality of Working
Life Centre Toronto, 1981. [34] L. T. Hosmer, Trust: The connecting link between organizational theory and philosophical ethics, Academy of management Review 20 (1995) 379–403. [35] A. Deshpande, H. Sharp, Responsible ai systems: who are the stakeholders?, in: Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society, 2022, pp. 227–236. [36] C. D. Wirz, J. L. Demuth, A. Bostrom, M. G. Cains, I. Ebert-Uphof, D. J. Gagne II, A. Schumacher, A. McGovern, D. Madlambayan, (re) conceptualizing trustworthy ai: A foundation for change, Artificial Intelligence (2025) 104309. [37] J. Webster, R. T. Watson, Analyzing the past to prepare for the future: Writing a literature review, MIS quarterly (2002) xiii–xxiii. [38] U. Wilkens, D. Lupp, V. Langholf, Configurations of human-centered ai at work: seven actor-structure engagements in organizations, Frontiers in Artificial Intelligence 6 (2023) 1272159. [39] Z. Wang, C. Huang, X. Yao, A roadmap of explainable artificial intelligence: Explain to whom, when, what and how?, ACM Transactions on Autonomous and Adaptive Systems 19 (2024) 1–40. [40] N. M. Pless, A. Sengupta, M. A. Wheeler, T. Maak, Responsible leadership and the reflective ceo: Resolving stakeholder conflict by imagining what could be done, Journal of Business Ethics (2021) 1–25. [41] T. K. Mitchell, M. Popa, R. E. Ashcroft, S. Prasad, A. Sharp, C. Carnforth, M. Turner, A. Khalil, N. Fenwick, S. Leven, et al., Balancing key stakeholder priorities and ethical principles to design a trial comparing intervention or expectant management for early-onset selective fetal growth restriction in monochorionic twin pregnancy: Fern qualitative study, BMJ open 14 (2024) e080488. [42] M. Langer, D. Oster, T. Speith, H. Hermanns, L. Kästner, E. Schmidt, A. Sesing, K. Baum, What do we want from explainable artificial intelligence (xai)?–a stakeholder perspective on xai and a conceptual model guiding interdisciplinary xai research, Artificial intelligence 296 (2021) 103473. [43] M. Langer, C. J. König, Introducing a multi-stakeholder perspective on opacity, transparency and strategies to reduce opacity in algorithm-based human resource management, Human Resource Management Review 33 (2023) 100881. [44] M. Kinney, M. Anastasiadou, M. Naranjo-Zolotov, V. Santos, Expectation management in ai: A framework for understanding stakeholder trust and acceptance of artificial intelligence systems, Heliyon 10 (2024). [45] M. Kim, S. Kim, J. Kim, T.-J. Song, Y. Kim, Do stakeholder needs difer?-designing stakeholder-tailored explainable artificial intelligence (xai) interfaces, International Journal of Human-Computer Studies 181 (2024) 103160. [46] A. Kawakami, D. Wilkinson, A. Chouldechova, Do responsible ai artifacts advance stakeholder goals? four key barriers perceived by legal and civil stakeholders, in: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, volume 7, 2024, pp. 670–682. [47] S. Haas, K. Hegestweiler, M. Rapp, M. Muschalik, E. Hüllermeier, Stakeholder-centric explanations for black-box decisions: an xai process model and its application to automotive goodwill assessments, Frontiers in Artificial Intelligence 7 (2024) 1471208. [48] M. Marzouk, C. Zitoun, O. Belghith, S. Skhiri, The building blocks of a responsible artificial intelligence practice: An outlook on the current landscape, IEEE Intelligent Systems 38 (2023) 9–18. [49] S. Gupta, R. Jaiswal, How can we improve ai competencies for tomorrow’s leaders: Insights from multi-stakeholders’ interaction, The International Journal of Management Education 22 (2024) 101070. [50] R. R. Hofman, S. T. Mueller, G. Klein, M. Jalaeian, C. Tate, Explainable ai: roles and stakeholders, desirements and challenges, Frontiers in Computer Science 5 (2023) 1117848. [51] R. Madhavan, J. A. Kerr, A. R. Corcos, B. P. Isaacof, Toward trustworthy and responsible artificial intelligence policy development, IEEE Intelligent Systems 35 (2020) 103–108. [52] S. Fukuda-Parr, E. Gibbons, Emerging consensus on ‘ethical ai’: Human rights critique of stakeholder guidelines, Global Policy 12 (2021) 32–44. [53] A. Kasirzadeh, Reasons, values, stakeholders: A philosophical framework for explainable artificial intelligence, arXiv preprint arXiv:2103.00752 (2021). [54] C. Figueras, C. Rossitto, T. Cerratto Pargman, Doing responsibilities with automated grading systems: An empirical multi-stakeholder exploration, in: Proceedings of the 13th Nordic Conference on Human-Computer Interaction, 2024, pp. 1–13. [55] D. Domínguez Figaredo, J. Stoyanovich, Responsible ai literacy: A stakeholder-first approach, Big Data & Society 10 (2023) 20539517231219958. [56] N. R. Haddaway, M. J. Page, C. C. Pritchard, L. A. McGuinness, Prisma2020: An r package and shiny app for producing prisma 2020-compliant flow diagrams, with interactivity for optimised digital transparency and open synthesis, Campbell systematic reviews 18 (2022) e1230. [57] M. M. Nasr-Azadani, J.-L. Chatelain, The journey to trustworthy ai-part 1: Pursuit of pragmatic frameworks, arXiv preprint arXiv:2403.15457 (2024). [58] World Economic Forum, 4 steps to developing responsible ai, 2019. URL: https://www.
weforum.org/agenda/2019/06/4-steps-to-developing-responsible-ai/, accessed: 2025-04-06. [59] A. B. Arrieta, N. Díaz-Rodríguez, J. Del Ser, A. Bennetot, S. Tabik, A. Barbado, S. García, S. Gil-López, D. Molina, R. Benjamins, et al., Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai, Information fusion 58 (2020) 82–115. [60] M. Bovens, Analysing and assessing accountability: A conceptual framework 1, European law journal 13 (2007) 447–468. [61] V. Dignum, Responsible artificial intelligence: designing ai for human values (2017). [62] R. V. Zicari, J. Brusseau, S. N. Blomberg, H. C. Christensen, M. Cofee, M. B. Ganapini, S. Gerke, T. K. Gilbert, E. Hickman, E. Hildt, et al., On assessing trustworthy ai in healthcare. machine learning as a supportive tool to recognize cardiac arrest in emergency calls, Frontiers in Human Dynamics 3 (2021) 673104. [63] A. W. Flores, K. Bechtel, C. T. Lowenkamp, False positives, false negatives, and false analyses: A rejoinder to machine bias: There’s software used across the country to predict future criminals. and it’s biased against blacks, Fed. Probation 80 (2016) 38. [64] Government of Japan, Social Principles of Human-Centric AI, Technical Report, 2019. URL: https://www.cas.go.jp/jp/seisaku/jinkouchinou/pdf/humancentricai.pdf.