The new ethical theories that take the distributed agency into account can help advance AI ethics. Conventional moral frameworks address individuals, and human responsibility assigns rewards or punishments according to personal choices and intentions. However, dispersed agency suggests that all players share accountability, which is the case with AI and, for instance, with firms, customers, software/hardware, designers, and developers (Taddeo and Floridi, 2018).

The most well-known examples are the many utilitarian perspectives that date back to Bentham (1789) and Mill (1861). They are predicated on the notion that, in theory, it is possible to total the benefits and drawbacks of a specific course of action. The ethically best choice is the one that has the maximum net utility, or utility less disutility. Theories of utilitarianism, often known as “consequentialist” or “teleological,” describe moral behavior that aims to maximize the “good” for the majority (Starr, 1983; McGee, 2010; Gustafson, 2013).

On the other hand, deontology is predicated on the idea that an action’s ethical assessment begins with the agent’s obligation while carrying it out. “Act only on that maxim by which you may at the same time will that it should become a universal law” is the most frequently cited expression of the categorical imperative (translation, given in Bowie, 2017). This categorical imperative prevents actors from justifying their own exemptions. Deontological theories, which have given rise to business social responsibilities (Mazutis, 2014; Lin et al., 2022), aim to explain moral behaviors as a set of standards or codes of conduct. Deontological ethics is still criticized today, usually for being overly strict and for potential contradictions between obligations. For instance, conflicting moral commitments might create situations in which following the rules to the letter may have unfavorable effects (O'Neill, 1989).

According to virtue theories, people behave in accordance with their inner “moral compass” when taking activities (White et al., 2023; James et al., 2023; Grant and McGhee, 2022). Some academics contend that deontology does not provide a workable framework for addressing moral disputes in everyday situations (Darwall, 2009). Table 1 summarizes key features of each ethical theory.

The term “responsible AI governance” refers to the systems put in place by businesses to deal with the moral questions raised by AI. Some of the most important ideas in responsible AI governance that have been discussed in the literature are stakeholder involvement, openness, justice, and accountability. According to Zhang and Lu (2021), these topics are essential for resolving public worries over prejudice, discrimination, and the possible abuse of AI systems.

Integrating responsible AI concepts into governance frameworks is crucial, according to a systematic review by Batool et al. (2023). To reduce the dangers of AI deployment and make sure that ethics are part of AI development from the start, this integration is crucial. There is a notable void in academic research and practical implementation, as the review finds no comprehensive frameworks that fully address these ideas.

Finding out who is responsible for AI system decisions is a major issue in AI governance. Organizational executives and legislators should also be involved in holding developers accountable (Morley et al., 2023). In order to gain the public’s trust, AI systems must be transparent. It is important for stakeholders to comprehend the decision-making process of AI systems. Users can understand the reasoning behind automated judgments with the help of explainable AI tools (Camilleri, 2024). Equal treatment of AI results is another important issue. A number of studies have brought attention to the possibility that algorithmic bias could cause discrimination against vulnerable populations (Batool et al., 2023). It is crucial to involve varied stakeholders, including affected communities, ethicists, and legal experts, when building frameworks for inclusive governance. The values and aspirations of society toward AI technology can be better understood with the help of stakeholder involvement (Mäntymäki et al., 2022).

The concept of corporate governance has evolved significantly over the years, leading to the development of various models that reflect different cultural, economic, and legal contexts. The most prominent models include the Anglo-American model, the Continental European model, and the Asian model. Each of these models presents unique features that influence corporate behavior and stakeholder relationships.

A comparison shows how different governance models handle agency theory, conflicts between shareholders (principals) and management (agents). Performance-based compensation matches managerial incentives with shareholder interests under the Anglo-American paradigm (Grove et al., 2011), but it may overlook stakeholder concerns. The Continental European model’s dual-board structure provides additional scrutiny but may slow decision-making (Vig and Datta, 2018).

In these frameworks, emerging studies emphasize corporate social responsibility (CSR). Kaur and Singh (2018) discovered that organizations with excellent governance frameworks are more likely to engage in CSR, improving their long-term performance. Technology transforms corporate governance, according to recent research. Digital tools improve data reporting and stakeholder communication, increasing transparency (Adams et al., 2010).

The most often stated ethical principles in AI ethics standards are transparency, privacy, accountability, and fairness, according to a thorough literature assessment by Khan et al. (2022). These tenets form the cornerstone of legislative frameworks aimed at efficiently regulating AI technologies. A complete regulatory approach is demonstrated by the AI Act proposed by the European Commission, which creates a legislative framework that classifies AI systems according to risk categories. Before being introduced into the market, high-risk AI systems have to adhere to stringent safety, transparency, and accountability regulations (Kargl et al., 2022).

The High-Level Expert Group on AI created the Ethics Guidelines for Responsible AI in 2019, which outline seven more essential criteria that AI systems must fulfill: accountability, diversity, non-discrimination and fairness, privacy and data governance, human agency and oversight, technological robustness and safety, and transparency (Palumbo et al., 2024). The purpose of these guidelines is to establish a common norm for moral AI practices among EU member states.

AI adoption involves ethical, regulatory, and social responsibility. As AI is implemented, companies must manage significant ethical issues. AI deployment decisions require ethical frameworks, according to research. Binns (2018) suggests that AI system implementation should be transparent, fair, and accountable. A thorough literature study by Rjab et al. (2023) found that smart cities need ethical rules to mitigate AI technology dangers.

Organizational culture also influences responsible AI adoption. Shahzadi et al. (2024) found that ethical companies prioritize responsible AI activities. This includes encouraging employees to raise ethical and bias issues regarding AI systems. Organizational culture and accountability affect how AI technologies are perceived and used across sectors. According to the literature, organizations should actively work with lawmakers to create AI-specific rules. Organizations can encourage innovation and public interest by participating in regulatory framework discussions (Madan and Ashok, 2023).

Responsible AI requires stakeholder participation. Sharma (2024) advises stakeholders to address AI’s ethical effects on human interactions and society. The author claims that AI systems change human interactions and ethical norms as they become more incorporated into daily life. AI in transportation or healthcare can revolutionize how people interact with each other and technology. This requires parties to work together to prevent AI from undermining human values and social cohesion.

AI transparency and accountability needs vary by stakeholder. Hind et al. (2019) found that regulators want AI systems to be fair and safe, whereas end-users want explanations to develop trust and improve decision-making. This divergence emphasizes the need for stakeholder-specific communication techniques. Weller (2019) also notes that developers need system performance data to debug and improve, demonstrating the interdependence of stakeholder roles in responsible AI implementation.

For stakeholder decisions, responsible AI frameworks are needed. Deshpande and Sharp (2022) list individuals, organizations, and international bodies as responsible AI system stakeholders. Their research implies that collaborative ethical rules can improve AI responsibility at all levels. Stakeholder viewpoints in AI governance can reduce technological misuse concerns and benefit society. Figure 1 focuses on ethical frameworks but does not encompass broader organizational enablers like culture or leadership. These factors are addressed separately in Section 4 as part of the implementation strategy layer.

Algorithm and model ethics covers topics such as machine decision-making, algorithm selection processes, training and testing of AI models, transparency, interpretability, explainability, replicability, algorithm bias, error risk, and transparency of data flow. Predictive analytics ethics covers topics such as discriminatory decisions and contextually relevant insight. Normative ethics covers topics such as bias by generalizing AI conclusions, justice, fairness, and inequality. Relationship ethics covers topics such as user interfaces and human-computer interaction, as well as relationships between patients, doctors, and other healthcare stakeholders (Saheb et al., 2021).

There are publications that cover models, frameworks, and methodologies that AI developers can use to improve their AI ethical implementation. For example, Vakkuri et al. (2021) provide the AI maturity model for AI software. There are further publications that describe the toolbox to handle fairness in ML algorithms (Castelnovo et al., 2020) and the transparency model to develop transparent AI systems (Felzmann et al., 2020).

Pant et al. (2024) grounded theory literature evaluation provides an understanding of practitioners’ viewpoints on AI ethics. This study shows that there is not a single, widely agreed-upon concept of AI ethics, which makes practical application difficult. The results imply that in order to guarantee that AI products function ethically and within society standards, ethical issues must be incorporated into the development stages of the products. Five categories, awareness, perception, need, difficulty, and approach, are identified in the research in relation to practitioners’ experiences with ethics in AI.

Among the various applied ethics models, decision-making frameworks occupy a central role in guiding how ethical judgments are made during AI development. The next subsections analyze multiple ethical decision-making models and outline how such models support dynamic risk governance throughout the AI lifecycle.

New risks arise as AI develops, necessitating flexible management techniques. An AI Risk Management Framework that highlights trustworthiness factors throughout the AI lifecycle has been proposed by the National Institute of Standards and Technology (NIST). With the potential to both create new and worsen pre-existing problems, this paradigm seeks to assist enterprises in navigating the intricacies of risk management related to generative AI technologies (AI N, 2024). The process of assessing ethical concerns in AI systems, which includes their detection, appraisal, and mitigation, is shown in Figure 3.

Although traditional enterprise risk management (ERM) frameworks offer structure, they are too static for the dynamic and unpredictable nature of AI. For example, they often fail to capture emergent risks like algorithmic discrimination or deepfake misuse, which require real-time ethical oversight and iterative updates, a need rarely addressed in current ERM implementations (Olson and Wu, 2017; Baquero et al., 2020). Their strategy consists mostly of the following steps: risk analysis (risk identification), risk assessment, risk management, and risk control. Such models operate in a very static manner, making them unsuitable for dynamic purposes such as AI (Baquero et al., 2020).

Table 3 outlines key steps in designing a risk management framework for industrial AI, including identifying and prioritizing AI use cases (Stuurman and Lachaud, 2022; Greiner et al., 2022) and evaluating existing frameworks (Butcher and Beridze, 2019). It emphasizes risk quantification (Bannister and Connolly, 2020), legal and regulatory considerations (Mäntymäki et al., 2022), and maintaining AI safety and reliability (Nikitaeva and Salem, 2022). The framework should also support innovation and enable dynamic algorithm regulation (de Almeida et al., 2021) while aligning with evolving legal standards.

Research suggests that the development of exhaustive ethical frameworks that are specifically tailored to specific contexts can be achieved through self-regulatory efforts. Numerous organizations have established ethical guidelines that address critical issues, including algorithmic fairness, user autonomy, and data privacy (Tahaei et al., 2023). Nevertheless, critics contend that self-regulation may not possess the rigor and enforcement mechanisms required to guarantee industry-wide compliance. The absence of external oversight can lead to the inconsistent application of ethical standards and may enable companies to prioritize profit over ethical considerations (Giarmoleo et al., 2024). Advocates for government regulation contend that it is indispensable for safeguarding fundamental rights and guaranteeing accountability in AI systems, particularly those classified as “high-risk” (Khan et al., 2022).

Seven fundamental ethical precepts that AI systems must follow in order to be trusted were found through a thorough literature review: Social and Environmental Well-Being, Technical Robustness and Safety, Privacy and Data Governance, Transparency, Diversity, Non-Discrimination and Fairness, Human Agency and Oversight, and Accountability in the workplace. The High-Level Expert Group (HLEG A, 2019) on AI stated these values. The emphasis on these ideas highlights how important it is for AI systems to empower people while guaranteeing their transparent and safe operation. The difficulty is in creating objective metrics to evaluate adherence to these guidelines because subjective measurements might introduce biases and inconsistencies into ethical analyses.

Adoption of AI technologies depends heavily on trust, especially in delicate industries like finance and healthcare. A thorough analysis of the moral concerns surrounding large language models (LLMs) highlights how crucial it is to remedy inadvertent injuries, maintain openness, and adhere to human values. The report promotes a multipronged strategy that includes public involvement, industry accountability, regulatory frameworks, and ethical oversight. In order to ensure that ethical issues are integrated throughout the machine learning process, this collaborative effort is crucial for changing norms around AI development (Ferdaus et al., 2024).

There is a pressing need for standardized practices, as evidenced by the emergence of AI ethics guidelines across the globe. Significant differences in ethical principles between different jurisdictions were found in a meta-analysis of 200 governance regulations, underscoring the difficulty of creating standards that are applicable to all situations (Corrêa et al., 2023). Effective governance and accountability in AI systems may be hampered by this mismatch. A cohesive framework that may direct organizations in implementing moral practices while taking local conditions into account is called for by the review (Corrêa et al., 2023).

AI control systems are meant to monitor the actions of autonomous agents and hold them to ethical boundaries. LawZero’s approach is a good model of such a method, which seeks to build systems that can monitor and intervene in real-time when necessary. Tamang and Bora (2025) Enforcement Agent (EA) Framework is a good case in point, including supervisory agents within worlds for the purpose of monitoring and correcting other agents’ action.

The ethical consequences of autonomous systems are profound, as Amoroso and Tamburrini (2019) highlight. They touch on the necessity of human control in the operational autonomy of robot systems, particularly in high-stakes applications such as military endeavors and medicine. The challenge lies in finding a balance between the benefits of autonomy and the requirements of accountability and ethical intervention. With very advanced AI systems, the risk of misalignment with human values increases, which necessitates robust ethical structures.

Research on the design of responsible AI is essential because it deals with the nexus of technology, ethics, and society influence. The “Ethics by Design for AI” (EbD-AI) framework is a methodical way to include ethical issues into AI development. This concept places a strong emphasis on how fundamental moral principles, like liberty, privacy, justice, openness, responsibility, and well-being, should be included into AI systems at every turn of their existence. This method has been incorporated by the European Commission into its ethics evaluation processes for AI projects, emphasizing how important it is to make sure that moral values are operationalized in real-world applications rather than just being theoretical (Brey and Dainow, 2023).

A growing number of international organizations have released guidelines in response to the growing interest in AI ethics. Jobin et al. (2019) conducted a research that yielded 84 documents that delineated ethical criteria for AI. The study revealed a convergence around several core values, including responsibility, justice, transparency, and fairness. These recommendations are in keeping with the increasing consensus about the necessity of ethical frameworks that oversee AI research while also guaranteeing adherence to the law and social norms.

One of the primary limitations of existing responsible AI frameworks is the lack of appropriate evaluation mechanisms. Reddy et al. (2021) highlight that AI systems in healthcare have been developed without a comprehensive evaluation of their translational aspects, such as functionality, utility, and ethics. The Translational Evaluation of Healthcare AI (TEHAI) framework was proposed to address these gaps by focusing on capability, utility, and adoption. However, the limited focus of most existing frameworks on reporting and regulatory aspects risks overlooking the day-to-day application of ethical principles in practice.

The ethical challenges associated with AI use are particularly acute in specialist areas such as palliative care. De Panfilis et al. (2023) discuss the ethical issues of AI-based clinical decision-making systems in palliative care, advocating for an equilibrium position that honors patient autonomy, quality of life, and psychosocial context of care. AI application for mortality prediction, while beneficial, is concerning due to the potential for oversimplification in complex decision-making. This mirrors a broader issue in which ethical frameworks may be unable to address the specific challenges posed by AI across a range of healthcare environments.

Doyen and Dadario (2022) outline several technical and operational pitfalls that are to blame for the low clinical impact of AI technologies. These include pitfalls related to data quality, algorithmic bias, and the failure to integrate AI systems with existing workflows. The authors argue that without addressing such underlying issues, ethical frameworks alone cannot ensure effective implementation of AI in healthcare. The need for an integrated approach incorporating both ethical and practical implementation aspects is necessary for overcoming such obstacles.

Another important consideration in the adoption of AI tools in healthcare is their trustworthiness. Lekadir et al. (2025) highlight the need to establish trust between patients, clinicians, and healthcare organizations. The FUTURE-AI guideline, which was developed by international consensus, provides best practices for developing responsible AI tools. The challenge, however, lies in translating these guidelines into effective practice since trust would typically be established with repeated and transparent interactions with AI systems.

The influence of stakeholders on the development and implementation of responsible AI frameworks is a major concern. Heymans and Heyman (2024) argue that existing guidelines have a tendency to codify the agendas of powerful stakeholders rather than the interest of the broader public. This can lead to ethical frameworks that are disconnected from practical situations, undermining their usefulness. A more extensive approach that involves diverse stakeholder perspectives is needed to create ethical guidelines that are grounded in the realities of AI adoption.

Medical education is another area where ethical frameworks are confronted with the introduction of AI. Ma et al. (2024) stress overall AI literacy among medical students as a precursor to secure and responsible AI-assisted patient care. Education today tends to focus more on the ethical aspects than the technical proficiency, and as a result, there exists a knowledge gap in assessing and deploying AI technologies in clinical practice. Such educational gaps are essential to be addressed to develop a workforce capable of handling the intricacies of AI in healthcare.