Algorithmic governance and criminal justice: a comparative analysis of ethical AI frameworks in Jordan and Oman

Abstract

Background:

The increasing application of artificial intelligence (AI) within criminal justice poses problems related to transparency, accountability, and human rights. Although the international discourse on ethical AI has gained attention, comparative research in Global South settings is still scarce.

Methods:

This paper employs qualitative comparative analysis to explore AI governance in Jordan and Oman’s criminal justice systems. It examines policy texts covering the period from 2020 to 2025, focusing on principles of transparency, accountability, fairness, privacy, and human oversight, guided by the “silicon cage” framework.

Results:

Both Jordan and Oman align with global ethical AI standards; however, a policy -law gap persists, as governance frameworks rely primarily on non-binding instruments. While similar principles are present, Jordanian policies emphasize human rights, whereas Omani policies prioritize development-oriented objectives.

Conclusion:

Convergence in normative principles does not translate into uniform regulatory implementation. Contextual factors play a decisive role in shaping governance outcomes.

1 Introduction

The rapid incorporation of artificial intelligence technology in various public institutions has marked one of the most profound changes in the current governance structure. In various nations around the world, algorithmic technologies are being used to improve the decision-making processes of various public institutions, the police, and the judicial system. This can be viewed as an extension of the current changes that are taking place in the governance structure, also known as algorithmic governance, where various computational technologies are being used to improve the interactions between the government and its citizens. In this new form of governance, the conventional bureaucratic governance, which has been based on conventional legal principles, administrative practices, and institutional responsibilities, has been slowly moving towards what has been described as algorithmic rationality, where various forms of decision-justification are taking place (Criado et al., 2025; Gao and Luna-Reyes, 2025).

This change can be viewed as part of the broader socio-technical change that has been described as the “datafication” of society, where various forms of social interactions, institutional practices, and governance processes have been translated into digital data and computational systems. Although this has provided greater efficiency, predictability, and optimization of various administrative processes, this change has also raised critical questions regarding the governance of the public sphere, democratic governance, and the protection of human rights (Babšek et al., 2025; Robles and Mallinson, 2025). There have been increased concerns that this form of governance might change the conventional relationship between the government and its citizens by incorporating various forms of automated authority that are based on computational systems, as opposed to conventional administrative systems (Alnemr, 2024; Elliott and MacCarthaigh, 2025).

The issues listed above are especially important when we think about the impact on the criminal justice system because these issues directly affect an individual’s liberty, their right to procedural justice, and ultimately the legitimacy of the legal system.

As an illustration of this concern, if a criminal court uses a predictive model (algorithm) to determine the likelihood that a defendant will commit another crime prior to trial so that the court can make informed decisions regarding pretrial detention, it is likely that the reason behind the decision made by the predictive model will be unavailable to both judges/judicial officials and defendants due to either copyright protection, or the complexity of technology. This illustrates why there is an immediate need to examine how algorithmic decision-making affects the relationship between government and governmental agencies (in terms of their responsibility to provide transparency), and the publics’ perceptions of governmental responsibility to ensure transparency.

In the developing body of research into algorithmic governance, such processes have been conceptualized as an integral component of a more overarching “silicon cage” of digital governance, wherein digital infrastructures are increasingly seen to play an influential role in the governance and production of knowledge within society. It is therefore the case that, while algorithmic systems are seen to play an important role in automating administrative tasks, they are also instrumental in the production of the information environment within which such decisions are made and justified. Increasing reliance on algorithmic systems for decision support has raised concerns over the crystallization of existing social asymmetries, as reflected within sensitive institutional domains such as criminal justice (Borinca, 2025; Hannah-Moffat, 2019; Zajko, 2021).

In terms of empirical research, studies into predictive policing and risk assessment have highlighted the potential for algorithmic systems to perpetuate structural inequalities, unless such fairness is built into the governance and design of such systems (Alikhademi et al., 2022; Farayola et al., 2025).

Closely related to the problem of bias is the problem of algorithmic opacity, or the so-called black box problem. It has been noted that machine learning algorithms, especially sophisticated ones, can be based on complex statistical operations that are not always easy to understand, even for the creators of the algorithms. This problem can be especially critical for legal systems that stress the importance of transparent ‘reasoning’ and procedural accountability. In this regard, it has been argued that if legal decisions are based on complex algorithmic operations that are not transparent enough, the very legitimacy of legal decisions can be challenged by legal practitioners and ordinary citizens alike (Rajendra and Thuraisingam, 2025; Sode, 2026). In this regard, algorithmic decision systems also pose some broader questions regarding the compatibility of automated governance technologies with some of the most basic principles of democratic accountability and the so-called rule of law.

In this regard, the increasing tendency toward the application of AI systems in public governance has led to the creation of ethical and regulatory guidelines by international organizations and national governments that are aimed at promoting the proper application of algorithmic technologies. The UNESCO Recommendation for the Ethics of Artificial Intelligence and the newly emerging regulatory guidelines, such as the European Union’s Artificial Intelligence Act, have emphasized the importance of recognizing that algorithmic systems, especially those applied for high-risk purposes, should remain transparent and be subjected to some kind of institutional accountability (Anastasopoulou, 2025; Coglianese and Crum, 2025). However, the creation of proper national systems of governance of AI systems remains a complex issue, especially for countries that are facing the problem of rapid digitalization.

The researcher will examine this concept further by conducting a comparative evaluation of how Jordan (the Hashemite Kingdom) and Oman (Sultanate) design and regulate AI. Both countries are developing digital transformation plans that could be very important for modernizing their governments. With each government undergoing modernization, they were required to create policy frameworks on using AI in an ethically responsible way. From an ethics standpoint, the first country to adopt a national charter to govern the use of AI was the Hashemite Kingdom of Jordan with its introduction of the National Charter for Artificial Intelligence Ethics in 2022. Conversely, the Sultanate of Oman adopted its general policy for the safe and ethical use of artificial intelligence systems in 2025 as part of its larger Oman Vision 2040.

Although there is an increasing number of studies examining algorithmic governance, there is still very little comparative knowledge about how the ethical aspects of AI are implemented in different forms of governance that govern the criminal justice system in countries of the Global South, with a particular focus on the Middle East. The study provides a comparison of Jordan and Oman, comparing the way in which global ethical standards for AI are translated into governance systems at the country level and how these governance systems represent varying institutional logic. Through investigating the manner in which artificial intelligence governance systems in both the Hashemite Kingdom of Jordan and the Sultanate of Oman were influenced by global ethics, the study will contribute to the ongoing debate about what form governance takes in the age of artificial intelligence.

2 Literature review and conceptual framework

2.1 Artificial intelligence and the digital transformation of governance

In this regard, artificial intelligence can be seen to have emerged as a major force in institutional changes in governance systems. Today, many governments across the world have been seen to use algorithmic technologies to facilitate their administrative decisions, improve their services to the people, and make their institutions more efficient. All these can be seen to have emerged as an integral part of the phenomenon of digital governance, where technology plays an important role in facilitating interactions between people and institutions (Criado et al., 2025; Gao and Luna-Reyes, 2025). While these studies emphasize efficiency and service delivery, they tend to underexplore the implications of such transformations for institutional accountability and legal oversight, particularly in high-stakes domains such as criminal justice.

It can be seen that artificial intelligence technology can be viewed not only as a technology but also as an integral part of socio-technical systems, law, politics, and institutions. In this regard, the technology can be seen to play an important role in helping institutions process large amounts of information, make predictions, and provide support to institutions in their automated decision-making processes. All these have significant implications for traditional governance systems, as they have been seen to bring in new actors, such as technology developers, data scientists, and technology firms, into the process (Elliott and MacCarthaigh, 2025). However, existing scholarship differs in its interpretation of these developments, with some viewing them as enhancing administrative capacity, while others highlight the risks of diffused responsibility and weakened institutional accountability.

The development of algorithmic systems can also be seen to have been associated with the phenomenon of datafication, which refers to the increasing tendency to convert social practices and administrative processes into digital data. All this can also be seen to have increased the potential of institutions to observe patterns, predict outcomes, and deal with complex issues in governance. Simultaneously, the implications of algorithmic governance for the nature of authority create additional challenges regarding transparency, accountability, and democratic legitimacy as well (Robles and Mallinson, 2025; Vatamanu and Tofan, 2025). The conflict between these two values, efficiency and legitimacy, is an open challenge within the body of research on this topic, especially with regard to environments where legal sensitivities may exist.

2.2 Societal implications of algorithmic decision-making

There is a great deal of controversy surrounding the role of Artificial Intelligence in governance. Algorithmic decision making technology has proven to increase efficiency of processes and results, yet many are concerned about the potential negative impacts of such technologies.

The most notable concern associated with AI technologies is bias. Since machine learning models utilize past data to predict future outcomes, they may also reflect past biases. Therefore, utilizing such models in an algorithmic system could potentially exacerbate inequality within society (Ahmad et al., 2026; Zajko, 2021). There are numerous areas where researchers have identified a risk of bias from algorithms, including credit lending (Farayola et al., 2025), hiring practices (Barocas and Selbst, 2016) and crime prediction. Early research focused on identifying bias, whereas current research seeks to develop mitigation strategies. However, due to lack of consensus regarding the best way to institutionalize those safeguards into governance structures, the majority of researchers continue to focus on developing new strategies to mitigate bias from AI systems.

Another area where there are issues with regard to the above issue, yet an additional problem as well, is the “black box” problem which is otherwise known as algorithmic opacity. Many contemporary machine learning applications rely on sophisticated statistical processes to generate their output. The reliance on these types of statistical processes create a black box effect that will limit the ability of individuals to understand the rationale behind or challenge the decisions made by the application to their rights (Rajendra and Thuraisingam, 2025; Sode, 2026). The scholarly debates have been split between the proponents of technical-based explanations to provide transparency into the decision-making process vs. those who advocate for institutional and/or legal-based remedies to address this issue which highlights a divide between the technology and the governance oriented perspectives.

The concerns outlined above are further indicative of the fact that algorithmic governance is related to wider normative questions regarding fairness, accountability and political influence in digital governance systems. Although there exists an opportunity for AI to improve administrative efficiency, this could have the result of changing how institutions use and justify their respective powers (Alnemr, 2024; Ghosh et al., 2025). Thus, algorithmic governance needs to be viewed as a structural change in the nature of authority, rather than merely as a new technology.

2.3 Artificial intelligence in criminal justice systems

Criminal justice systems arguably present one of the most sensitive areas in which artificial intelligence systems can operate. In this respect, various criminal justice systems have been using artificial intelligence systems to support different functions within their systems. For example, risk assessment systems use historical information concerning different criminal activities to predict the likelihood of recidivism or pre-trial misconduct (Alikhademi et al., 2022; Mukhamadieva et al., 2025).

The use of artificial intelligence systems in criminal justice systems, in particular, raises critical legal and ethical issues. For example, different researchers have raised questions concerning whether people affected by artificial intelligence systems have a right to access information concerning the algorithms used in generating the results used in the systems, especially when such results have a bearing in determining the outcome of criminal trials (Koshevaliska, 2026; Vidaki and Papakonstantinou, 2025). In essence, artificial intelligence systems may not allow people to defend their rights when such rights have been impacted in some way by artificial intelligence systems. The consistency with which these rights can be enforced is still unestablished among many countries as well as reflects the overall degree of regulation fragmentation in this area.

Artificial intelligence systems used in criminal justice systems will raise additional legal questions for example legal professionals may need to determine what kind of “expert opinion”, “technical assistance” or “advice” that an artificial intelligence system provides. Whether artificial intelligence systems should be admitted into evidence, and if so how reliable their results would be, represents perhaps one of the most important emerging legal issues in today’s legal systems (Kheiri et al., 2025; Sierocka, 2025). This uncertainty further illustrates the absence of clear legal doctrine for the integration of AI into the current framework of law regarding both standards of evidence and the requirements of due process.

2.4 Global AI governance and ethical frameworks

The emergence of artificial intelligence has prompted global and national organizations to create multiple initiatives focused on developing governance models for technologies, including region-specific efforts in the Middle East and North Africa (Trigui et al., 2024). These governance initiatives aim to promote the implementation of ethical governance principles that will provide an optimal foundation for the development and implementation of all forms of algorithmic applications.

A major initiative internationally has been the UNESCO Recommendation on the Ethics of Artificial Intelligence. This recommendation recognized that there is a requirement to develop and implement AI based upon the established ethical principles of transparency, accountability, fairness, privacy and human oversight (UNESCO, 2021).

The development of these ethics-based governance frameworks have helped raise awareness of the importance of aligning AI systems with the core principles of human rights and democracy. By using the ethic of transparency it was possible to identify the level of transparency of algorithms. Additionally, use of accountability allowed for identification and resolution of potential issues resulting from discrimination due to the use of algorithms (Floridi et al., 2018; Ikram et al., 2026). Although the models outlined above provide standards for good governance, the majority of them have no compulsory enforcement mechanisms. Consequently, it is argued that many of the governance models established by organizations are used as ’soft law’ with the intent of providing an example to follow rather than being enforced by law (Buczynski et al., 2022; Liang, 2026). The disparity between the intended goal of creating governance norms and the limited ability to enforce them presents a significant void within existing research on comparative governance.

2.5 Comparative governance and institutional adaptation

Compared to a single-country study, comparative research would offer additional insights on how various countries deal with the new technologies’ developing issue. The regulatory systems for new technologies do not develop independently from one another. They can take place through a model of policy diffusion and learning. Even so, past comparative studies indicate that when there is a transfer of policies, it is unlikely that all countries will have the same regulatory approaches. National institutional contexts will determine how international normative standards for global governance are interpreted and implemented.

In addition, comparative legal studies demonstrate that even though different countries may use comparable regulatory principles when addressing new technologies, their governance systems could vary greatly. There may exist differences among countries regarding institutional capacities, policy objectives and socio-political structures. Some countries tend to emphasize more heavily regulatory controls and human rights whereas other countries tend to emphasize more heavily technological innovations and economic development. In light of this divergence, comparative research provides the opportunity to assess both formal regulatory alignment and the structural institutional logic which produces these governance paths.

A comparative study will provide researchers the opportunity to examine how global governance models interact with the varying institutional models used in each country when assessing responses to challenges presented by developing technologies. As such, a comparative approach provides an ideal way to identify how common ethical frameworks create divergent governance practices across different national environments.

2.6 Conceptual framework: from iron cage to silicon cage

The study examines how AI can be used to govern institutions and uses the “silicon cage” concept (a derivative of Max Weber’s “iron cage”) to do so. Max Weber said that bureaucracy is a way of governing using a form of rational-legal authority and is characterized by being ruled by rules, having hierarchical accountability, and being predictable. Bureaucracies can limit individuals but they have legitimacy because they operate under procedures with built-in transparency and accountability.

Algorithmic governance creates a type of institutional rationality that is fundamentally different than the bureaucratic form of rational legal authority. The silicon cage represents an emerging form of institutional rationality where decision making is impacted by AI through data analysis, predictive analytics, and machine learning algorithms. The silicon cage differs from the bureaucratic model in its use of statistical inference and machine learning algorithms to analyze large amounts of data; whereas bureaucracies rely on clearly defined rules to make decisions.

In addition to differences related to transparency, the Silicon Cage differs from the bureaucratic model in other important respects:

Firstly, whereas the Silicon Cage lacks a transparent framework due to its reliance upon machine learning algorithms (i.e., unexplainable), as opposed to the Bureaucratic System that has a clear set of rules (with explanations) to govern it. The way this is measured within this research is by analyzing the extent to which the governance frameworks have provisions related to transparency; specifically those dealing with explainability and disclosure, relating to artificial intelligence (AI).

Secondly, in the Bureaucratic System, decisions were made using logical reasonings; whereas decisions are made in Algorithmic Systems using prediction based on statistical correlation within vast amounts of data. This second measurement examines how much of an influence data driven logic’s (vs. explicit legal reasoning standards) drive the decision making process within governance frameworks.

Thirdly, in both types of systems, accountability exists but the way it occurs differs. Within Algorithmic Systems, there are several parties accountable for each other i.e., developers, analysts, users, etc. As well as regulatory bodies; whereas in Bureaucratic Systems, accountability is hierarchical. This third measurement is examined through the examination of accountability mechanisms and the distribution of accountability among institutional actors.

Fourthly predictive governance replaces or supplements retrospective adjudication. Traditional legal institutions focus on evaluating conduct after it happens. Algorithmic systems are increasingly focused on preemptively managing future risk. The extent to which governance frameworks include predictive tools and risk-based decision making in criminal justice processes is how this dimension will be measured.

These changes suggest that artificial intelligence might represent more than just a incremental change in institutional authority structures, and how they are exercised. Ethical governance strategies can also be conceptualized as an attempt to establish normative constraints on algorithmic systems before the silicon cage becomes fully embedded in governance structures.

The current study uses the above approach to conceptually understand how national governance structures in Jordan and Oman regulate artificial intelligence within their respective criminal justice systems. Additionally, international ethical guidelines for artificial intelligence regulation are translated into domestic regulatory approaches. Therefore, the silicon cage framework provides not only a theoretical lens, but also an analytical tool to evaluate comparative governance structures between the two cases.

3 Methodology

This study uses a qualitative comparative research design to explore how ethical principles for artificial intelligence are incorporated in governance frameworks used in the regulation of artificial intelligence in criminal justice systems in Jordan and Oman. The study only focuses on governance frameworks used in these jurisdictions, which have recently adopted national strategies for artificial intelligence while pursuing digital transformation in their judicial systems.

3.1 Research design

The study used qualitative document analysis as a form of research. Document analysis is considered a qualitative method of research that involves the systematic study of policy documents, strategic plans, and other official documents to identify patterns, themes, and governance structures embedded in textual materials (Bowen, 2009). The document analysis method is considered appropriate for the study because governance structures in developing technologies, including AI, are primarily developed in policy documents, strategic plans, and regulatory guidelines, and not in case laws and regulations. Document analysis would assist in the systematic study of policy documents to identify major ethical principles, regulatory guidelines, and institutional mechanisms for the integration of AI technology in the criminal justice system. The comparative methodology that I use has two different dimensions: One dimension is the relationship between global ethical norms and national governance models; the other dimension is how institutions will be developed to apply global ethics to AI technology. Further, this research uses a comparative analytical model (i.e., comparing governance structures) across the two case studies to enable an understanding of how global norms are applied through national regulatory or governance structures.

3.2 Data sources

The information used for the study employs official policy documents and international ethical guidelines for the governance of artificial intelligence as tools. Official policy documents include international ethical guidelines for artificial intelligence, national strategic plans from government agencies concerning digital governance, and judicial strategies for Jordan and Oman.

An important reason for employing a policy document methodology is that such an approach will provide the opportunity to analyze how emerging technologies are governed through policy (as opposed to formal legislation).

Criteria were established for the collection of documents. These were (1) documents from internationally recognized organizations, or those from official government agencies; (2) relevance to artificial intelligence governance, digital transformation, or administrative/judicial systems; (3) date of issue within the time frame 2020–2025 to establish contemporaneity; and (4) availability in English or Arabic.

Following these criteria, purposive sampling was employed to collect policy documents related to AI governance developed by government agencies or international organizations during 2020–2025 which were written in either English or Arabic.

Purposive sampling was chosen so that only documents directly related to frameworks of governance and principles of ethics in artificial intelligence could be collected. This provided a high level of focus for analysis in relation to the purpose of the research.

Based on the selection criteria, eight policy documents were finally selected for Table 1.

3.3 Analytical procedure

A code was created for this project using a combination of deductive (top-down) and inductive (bottom-up) thinking. The deductive method referenced the ethical guidelines established through the UNESCO Recommendation on Ethics of Artificial Intelligence. For the inductive method, an initial close reading of the chosen documents took place. Following that close reading, additional work occurred on developing the indicator codes using terms found in governance frameworks.

The analysis will follow a thematic coding method that combines deductive coding based on previously identified ethical standards and inductive coding to identify emergent governance trends in the documents. Combining these two methods provides a balance between being theoretically relevant, while also providing sufficient empirical sensitivity.

The five organizing principles have been utilized as analytical constructs, as presented in Table 2.

The analysis proceeded in three stages. Stage 1 examined International Ethical Guidelines for Artificial Intelligence (AI) to evaluate the way they are represented in global governance structures. Stage 2 evaluated the representation of those international ethical guidelines in the National AI Governance Structures of Jordan and Oman. Stage 3 evaluated whether those guidelines are embedded within AI Governance Structures of Criminal Justice Institutions. Those guidelines could be embedded within those AI Governance Structures either directly or in the case of marginalization, indirectly through more general principles. Each stage was used consistently with each case study to provide comparative and analytical consistency when evaluating similarities and differences in governance structures.

To assess reliability, two researchers reviewed approximately thirty percent of all documents independently. Any inconsistencies between the two researchers regarding the interpretation of the data were addressed until both researchers had a common understanding of what constituted governance constructs. Inter-coder reliability was measured using Cohen’s Kappa Coefficient. That analysis indicated that there is an overall agreement between the two coders of κ=.82. This level of agreement indicates a high degree of coding reliability, supporting the consistency of the analytical process.

Due to the presence of some documents published in Arabic, all the Arabic documents were coded in their original language, as were the documents coded by the other researcher, as both have legal proficiency in the language. The regulatory terms were checked against available translations for consistency between languages. This level of agreement indicates a high degree of coding reliability, supporting the consistency of the analytical process.

3.4 Methodological limitations

The focus of this particular study was governance frameworks, as opposed to their empirical implementation. Therefore, it is possible to evaluate the normative and policy structures, as opposed to the practical application of AI in judicial processes. From the analysis, it was possible to see how ethical principles were articulated in policy documents, whether they were accompanied by institutional mechanisms, and so forth. However, it was not possible to evaluate whether they were implemented in practice or whether they influenced decision-making processes in institutions of criminal justice.

Another limitation of this study was that it was not possible to access informal governance, potential implementation issues, or actor perceptions that might impact the operation of the frameworks. This study can be built upon by further research that uses empirical approaches to evaluate the operation of governance frameworks in practice, as well as whether the policy commitment to ethical issues, as identified in the documents, translates into enforceable measures. Furthermore, the number of documents which were analyzed was limited and therefore it is difficult to generalize. In spite of this limitation, a comparison between the two case studies has been possible with respect to similarities as well as differences. Despite these limitations, the study provides a solid foundation for an analysis of structural and normative dimensions of governance of criminal justice systems by artificial intelligence.

4 Results and analytical findings

From the document analysis, it is possible to identify some patterns in the way ethical guidelines for artificial intelligence are implemented in governance systems in Jordan and Oman. Both systems show a high level of compliance with international ethical guidelines, and there are indications of differences in terms of institutional framing and the gap between policy and regulatory mechanisms. These findings are interpreted through the lens of the “silicon cage” framework, which enables an analytical understanding of how algorithmic governance reshapes transparency, reasoning, responsibility, and predictive authority within institutional contexts.

4.1 Normative convergence

The ethical standards developed in both Jordan and Oman are analogous to the international standards for AI governance. Both country’s domestic policy strategy provisions have similarities to the international policy strategy provisions for AI governance (disclosure, responsibility/accountability, fairness/justice, privacy/data security, etc.) The language of the domestic policy strategies in both countries also mirrors the language of international policy strategies for AI governance. These parallels reflect a policy diffusion process occurring, as well as an indication that the international standards for AI governance are key to creating national standards for AI governance. In this context, Jordan and Oman do not create their own national standards for AI governance. Instead, both nations use international standards for AI governance when developing their national standards for AI governance. From the viewpoint of the Silicon Cage framework, this convergence of norms reflects the diffusion of common transparent and accountable principles; however, the extent to which these principles are implemented varies across different institutional settings.

4.2 Divergent institutional logics

The comparison between the two countries shows distinct differences in the institutional logic (the way that an organization or institution thinks) that governs AI governance in both countries. The documents developed by Jordan have taken a rights-based approach to developing their AI governance, focusing on protecting the dignity of humans, their right to privacy, and ensuring they receive fair treatment through legal means within institutions of law. Documents developed by Oman connect the need for AI governance to its national ambitions to develop and modernize its society as outlined in Vision 2040. Here, ethical concerns are tied to promoting innovation at the national level, increasing efficiency in government operations, and utilizing technology in government processes. This distinction indicates how the different institutional logics influence the expression of “silicon cages”, where in Jordan, governance is focused on providing legal accountability and normative thinking; and in Oman, the focus is on more efficient and predictive governance with respect to the overall objective of achieving societal development.

4.3 The policy–legislation gap: soft governance and the limits of ethical charters

One of the major conclusions that can be made from the analysis is that policy instruments were used rather than regulations in the governance systems. The majority of the documents that were analyzed were strategic frameworks and ethical guidelines that outlined a normative position but did not impose a legal obligation.

While there is a certain level of freedom in governing dynamic technology with policy instruments, their non-binding nature might not be very effective in enforcing regulations and compliance with them. This gap between policy and legislation might suggest that existing governance might not be very effective in governing AI-assisted decision-making in criminal justice systems. Within the silicon cage framework, this gap reflects a transition toward distributed responsibility, where accountability is articulated normatively but lacks enforceable institutional mechanisms, thereby weakening hierarchical legal control.

4.4 Opacity and accountability: the challenge of algorithmic black boxes

The analysis indicates that there is a lack of emphasis with regard to the problem of algorithmic opacity. Although the two models emphasize transparency and accountability, the problem of algorithmic explainability is not addressed in the models.

Furthermore, in the criminal justice system, where algorithmic tools may be employed in decision-making processes in relation to risk assessment and investigation priorities, there are challenges in relation to procedural fairness and the right to dispute decisions made by algorithms due to a lack of explicit mechanisms for algorithmic explainability. This finding aligns directly with the silicon cage dimension of reduced transparency, where algorithmic reasoning remains opaque and limits the ability of affected individuals to understand or challenge decisions.

4.5 Stratification risks: algorithmic bias in criminal justice

The governance frameworks provide general references to fairness and non-discrimination, but they do not provide sector-specific measures to counterbalance the effects of algorithmic bias in a criminal justice setting. In fact, the documents do not provide concrete mechanisms to identify or reduce the effects of discrimination, which can be caused by algorithmic tools. This is a significant issue, as it is a known fact that algorithmic tools can, in fact, perpetuate social inequalities. The introduction of AI into the criminal justice system has the ability to exacerbate existing disparities due to the reliance on an algorithmically driven decision making process. The perspective of the Silicon Cage illustrates how statistically-based predictions will often replicate the same structural inequality patterns seen within society unless the CJ system incorporates safeguards in law.

To summarize, the data clearly shows that while AI governance frameworks for both Jordan and Oman exhibit significant normative alignment to international ethical standards they are primarily policy-driven. Therefore, despite ethical commitments being made at a strategic level; currently there are limited (if any) legal instruments available that could assist in addressing the complex challenges presented by algorithmically driven decision-making processes in CJ systems. Additionally, legal instruments would need to address the systemic issues created by AI and provide protections for citizens’ rights.

5 Discussion: governing the silicon cage

The research results from this study will contribute to a broader discussion about algorithmic governance and institutional authority within the digital era. In addition to showing a high degree of normative agreement among international AI ethics guidelines and national AI ethics agendas in both Jordan and Oman, the study’s findings suggest that this normative agreement does not necessarily translate into a greater degree of regulatory similarity. Rather, it illustrates how the ethical norms can provide for distinct logics of governance that independently shape understanding and implementation of AI as a part of each country’s respective national institutions.

At an empirical and theoretical level, the findings demonstrate that normative convergence occurs primarily at the level of discourse, whereas institutional translation remains context-dependent, as illustrated in Figure 1.

The figure highlights differences in institutional emphasis across transparency, reasoning, responsibility, and predictive governance dimensions.

5.1 Theoretical implications: beyond normative convergence

The study contributes to the growing literature on algorithmic governance that demonstrates how the spread of global ethical standards has impacted formal institutional similarities in ways that obscure deep institutional differences. Governance today is increasingly characterized by a type of change described by some as an evolution of the traditional social contract. Traditionally, the authority responsible for managing society was accountable to societal scrutiny. That accountability has been transferred to one or more computer-based algorithms, whose identities are currently unknown (Alnemr, 2024; Widłak, 2026). The silicon cage provides the context in which such a change represents an alteration in the nature of authority. Authority is no longer solely embedded in hierarchical structures and is increasingly being embedded in the algorithmic infrastructure.

In this case, Jordan and Oman’s adaptation strategies to a growing global trend are distinctly different. The discussion regarding governance in Jordan emphasizes a focus on a rights-based approach. Therefore, issues related to governance through artificial intelligence are viewed as a function of protecting human dignity; ensuring legal compliance with public institutions; and promoting procedural fairness. On the other hand, the Oman model views AI-governance through the lens of a larger strategy of modernizing administration and stimulating economic growth, as exemplified by Vision 2040. This demonstrates that while global policies may influence local policy-making, national institutional preferences, along with historical patterns of government, can also contribute to how we develop our algorithmic systems. Additionally, these differing models demonstrate how the “silicon cage” does not exist uniformly. Rather, it is formed based upon institutional logic which dictates whether an algorithmic system is first and foremost a legal risk object (as in the case of Jordan), or rather if it will be considered a developmental enabling infrastructure (such as in the case of Oman).

5.2 The silicon cage in criminal justice: distinctive stakes

The importance of algorithmic governance can be seen in the context of criminal justice institutions. While AI technology is being increasingly applied in various areas of public administration, the importance of algorithmic governance can be highlighted in the context of criminal justice institutions, where individual liberty and due process are at stake (Koshevaliska, 2026; Vidaki and Papakonstantinou, 2025). In terms of the silicon cage framework, criminal justice represents the most sensitive manifestation of algorithmic governance, as it directly affects core constitutional guarantees such as liberty, fairness, and procedural justice.

Given the importance of algorithmic governance, opacity in algorithmic decision-making raises serious normative concerns. While algorithmic decision-making can have serious implications in various areas of public administration, including judicial risk assessment, investigative decision-making, and analysis of evidence, the inability to interpret algorithmic decision-making can have serious implications for the right to explanation and the ability to challenge algorithmic decisions (Hassija et al., 2024; Rajendra and Thuraisingam, 2025). In this context, algorithmic opacity can be seen in the context of serious normative issues, including the compatibility of algorithmic decision-making with procedural justice. This directly corresponds to the transparency dimension of the silicon cage, where the opacity of algorithmic systems weakens traditional legal safeguards based on explainability and reason-giving.

Beyond the aforementioned limitations of transparency and explainability, a more fundamental jurisprudential limitation arises in the application of AI systems within criminal justice: their limited capacity to engage with the spirit of the law, rather than merely its textual formulation. Legal reasoning does not merely depend on a literal interpretation of legal provisions but may also require a purposive interpretation, judgment based on context, and an understanding of normative factors such as justice, equity, proportionality, etc. This illustrates an inherent contradiction that exists between the rationality of algorithms and the nature of legal interpretation. The “cage” created by these algorithms creates a preference for statistical inference as opposed to normative interpretations.

A “Spirit of the Law,” refers to the legal tradition in which legal meanings derive from broader sets of goals and values. Algorithmic systems, however, rely upon quantifiable data, statistical relationships, and formalized rule-based logic. Therefore, while algorithms may apply formal legal rules, they have no ability to engage with legal meaning in a larger or more holistic context. In this way, the limitations of algorithmic legal governance create an opportunity for judicial review based upon the necessity for human judgment within this area of criminal justice.

5.3 Comparative insights: institutional logics and governance pathways

The preceding comparative study has illustrated that the same ethical considerations can direct varying courses of action when interpreting their respective institutional logics regarding regulations. Both Jordan and Oman have incorporated international ethical standards as bases for their regulation, emphasizing transparency, accountability, fairness, and citizen participation; however, the overall policy context of each country’s regulation differs significantly. These differing narratives support the primary conclusion that algorithmic governance is not solely determined by technology but is instead an institutional mediation.

In the case of Jordan, the methodological approach of integrating AI governance into the existing legal and constitutional structures has placed a high level of emphasis upon protecting citizens’ rights through the judicial process. In contrast, the methodology for developing AI in Oman has emphasized its integration into a national modernization strategy which places innovation, administrative efficiency, and technological progress at its core. Therefore, this could lead to very different implications for future regulation. Rights-based governance emphasizes the development of law and accountability, while development-based governance focuses on institutional and technological innovations. Based on the silicon cage model, these two types of governance represent two contrasting methods of institutionalizing algorithms: one bound by legal rationality and the other facilitated by developmental rationality.

The results support previous research showing that contextual characteristics including institutional and administrative traditions have an influence on the ethical analysis of algorithmic governance (Criado et al., 2025; Gao and Luna-Reyes, 2025). This study expands upon the prior research by illustrating that the contextual variation also impacts how the four dimensions of the Silicon Cage, Transparency, Reasoning, Responsibility and Predictive Governance are prioritized and implemented.

5.4 Policy implications

Policymakers face challenges when trying to reconcile what is ideal about ethics with how they can be translated into policy tools. While both ethics and policy tools demonstrate the potential for high-minded ethics regarding AI policy, this research posits that the tools in place are primarily strategic policy tools. Therefore, it is no surprise to find there is a consistent disconnect between the articulated ethics and the legally binding enforcement of these ethics in the systems governing algorithms.

To make this leap, policymakers will require institutional tools to operationalize ethical values, e.g., regulations requiring transparency of algorithms, oversight bodies for AI systems utilized in government sectors, and procedural rights allowing individuals to contest automated decisions affecting their legal rights (Bignami, 2022; Coglianese and Crum, 2025). The above represent attempts to “unbox” the silicon cage by introducing accountability, oversight, and contestation into decision making processes based on algorithms.

In other words, we believe the regulation of AI should move past the symbolic ethics represented in existing systems. Governance scholars have made significant arguments that policymakers should establish AI policies that foster institutional innovations and social welfare through developing AI systems that are transparent, accountable, and ethical in their development and implementation (Ikram et al., 2026; Papagiannidis et al., 2025). Without institutionalizing these types of features, ethics frameworks may remain declarative, rather than transformational.

Therefore, from this analysis it can be argued that AI-governance within criminal justice systems should find an equilibrium point between technological innovation and legal accountability. Without stronger institutional tools; therefore, AI technological innovations could shift administrative efficiency into a new form of authority within modern governance systems.

This final note underscores the main claim of this research that the silicon cage is already functioning on at least some level, however, it continues to have potential for contestation with appropriate legal and institutional design.

5.5 Future research

Future Research Should Investigate How Practitioners Implement Ethical Principles in AI Governance Frameworks—Moving Beyond Document-Based Analysis.

The “silicon cage” framework has been developed empirically to describe the ways in which criminal justice institutions will implement ethical AI governance principles. However, empirical validation of the framework is still lacking. As such, researchers can investigate how each of the four dimensions of the framework (i.e., transparency, reasoning, responsibility, and predictive governance) manifest in real-world AI supported criminal justice systems. These investigations would provide an operational evaluation of algorithmic governance that goes beyond traditional normative policy analysis.

Comparative research should expand to include additional jurisdictions from the Global South. Comparing the implementation of AI governance across different jurisdictions will help researchers understand how differences in institutional environment influence how global ethics translate to national regulatory practices.

Finally, future research may want to consider using a mixed-method approach that includes legal analysis, practitioner interviews and/or case studies of AI deployments in courts or police departments to measure the disconnect between policy design and institutional practice.

6 Conclusion

This paper examined how ethical principles in artificial intelligence are embedded in governance structures for AI use in criminal justice systems through a comparative analysis of Jordan and Oman. The analysis revealed that Jordan and Oman share similar ethical principles in AI governance, including transparency, accountability, fairness, privacy, and human oversight, which are commonly found in international AI governance frameworks. However, this paper revealed differences in governance structures in Jordan and Oman. Jordan’s governance structure is based on rights protection and legal accountability in judicial bodies, while Oman’s AI governance structure is based on overall objectives for modernization in administration and economic development in line with Vision 2040. Such differences in governance structures of AI in Jordan and Oman illustrate how global ethical standards in AI use are shaped.

Moreover, this paper found a gap in terms of policy-legislation in AI governance in Jordan and Oman. Indeed, Jordan and Oman mostly depend on strategic policy documents and ethical guidelines in AI governance. However, there is a lack of legislation in Jordan and Oman that can be used to govern AI use in criminal justice systems. Even though AI governance frameworks in Jordan and Oman reveal a normative commitment to AI governance, there is a lack of legal power to ensure transparency, accountability, and procedural justice in AI use in criminal justice systems. The results of this research demonstrate an inherent weakness in present day governance models. These weaknesses occur when ethical commitments made by organizations are simply “stated” but do not have legal force or consequence.

This research contributes to the discussion on algorithmic governance by demonstrating that while normative convergence will lead to greater similarities in regulation, we see no such correlation. The evidence from this research suggests that the establishment of governance systems (including those for algorithmic) is dependent upon both the global standardization of ethics and the institutional logic of individual communities. As a result, the inclusion of technology/computational systems within existing public governance systems represents a microcosm of a larger phenomenon in which elements of institutional authority are being progressively mediated by technology. Thus, new forms of legal accountability arise due to the inclusion of technological systems within institutions.

Additionally, this research demonstrates that these issues can be studied and analyzed through the lens of the “Silicon Cage” model. Through this model, one can observe how various types of algorithmic systems affect transparency, reasoning, responsibility, and predictive authority in criminal justice governance.

Policymakers could utilize the findings from this research in multiple ways. For example, when considering AI governance in criminal justice, policymakers should consider moving from making ethical commitments towards establishing institutional frameworks based on those commitments. With respect to Jordan, examples of what this may look like include enacting legislation regarding AI in justice and law enforcement; creating auditing mechanisms that allow for the evaluation of AI; and providing educational/training programs for judges/judicial officials regarding algorithmic decision-making. This, in relation to Oman, might involve developing rights protection systems in relation to existing development-oriented strategies, developing systems of grievances for those who have been impacted by algorithmic decision-making, and developing systems of addressing artificial intelligence in national modernization initiatives. These recommendations underscore the need to transition from soft governance tools to regulatory frameworks that guarantee accountability and procedural justice.

In terms of areas for future research, this study offers a number of important insights. First, exploring how AI systems are implemented in judicial systems in Jordan and Oman might offer important insights into the practical implications of existing governance systems. Comparing this with other jurisdictions in the Global South might offer important insights into how national systems of governance respond to international ethical standards in local contexts. Finally, exploring how regional legal traditions and ethical frameworks impact AI governance in criminal justice systems might offer important insights for developing responsible AI governance systems in criminal justice systems. Future research would also benefit from empirical investigations involving judicial practitioners, policy implementers, and system developers in order to evaluate how ethical principles are operationalized in practice beyond policy-level commitments.

In conclusion, this study shows the need to move beyond symbolic ethical commitments to the development of systems that facilitate accountability and fairness, as AI systems become more embedded in the institutional structure of criminal justice governance systems. In this regard, the research shows that the phenomenon of algorithmic governance is not simply a development of technology but a change in the nature of institutional power itself, which can be observed via the “silicon cage” approach.

References

• AhmadA.VallèsY.IdaghdourY. (2026). Bias in AI systems: integrating formal and socio-technical approaches. Front. Big Data8, 1686452. 10.3389/fdata.2025.1686452

• AlikhademiK.DrobinaE.PrioleauD.PurvesD.GilbertJ. E. (2022). A review of predictive policing from the perspective of fairness. Artif. Intell. Law30, 1–17. 10.1007/s10506-021-09286-4

• AlnemrN. (2024). Democratic self-government and the algocratic shortcut: the democratic Harms in algorithmic governance of society. Contemp Polit Theory23, 205–227. 10.1057/s41296-023-00656-y

• AnastasopoulouM. (2025). Exploring algorithmic governance: the AI act and new realities for criminal justice and fundamental rights. New J. Eur. Crim. Law16 (2), 176–196. 10.1177/20322844251338627

• BabšekM.RavšeljD.UmekL.AristovnikA. (2025). Artificial intelligence adoption in public administration: an overview of top-cited articles and practical applications. AI6 (3), 44. 10.3390/ai6030044

• BarocasS.SelbstA. D. (2016). Big data’s disparate impact. SSRN. Electron. J. 10.2139/ssrn.2477899

• BignamiF. (2022). Artificial intelligence accountability of public administration. Am. J. Comp. Law.70 (Supplement_1), i312–i346. 10.1093/ajcl/avac012

• BorincaI. (2025). AI As moral cover: how algorithmic bias exploits psychological mechanisms to perpetuate social inequality. Anal. Soc. Issues. Public. Policy.25 (3), e70031. 10.1111/asap.70031

• BowenG. A. (2009). Document analysis as a qualitative research method. Qual. Res. J.9 (2), 27–40. 10.3316/QRJ0902027

• BuczynskiW.SteffekF.CuzzolinF.JamnikM.SahakianB. (2022). Hard law and soft law regulations of artificial intelligence in investment management. Camb. Yearbook Eur Legal Stud24, 262–293. 10.1017/cel.2022.10

• CoglianeseC.CrumC. R. (2025). Leashes, not guardrails: a management-based approach to artificial intelligence risk regulation. Risk. Anal.45, 4397–4407. 10.1111/risa.70020

• CriadoJ. I.Sandoval-AlmazánR.Gil-GarciaJ. R. (2025). Artificial intelligence and public administration: understanding actors, governance, and policy from micro, meso, and macro perspectives. Publ. Pol. Admin.40 (2), 173–184. 10.1177/09520767241272921

• ElliottM. T. J.MacCarthaighM. (2025). Accountability and AI: redundancy, overlaps and blind spots.” public performance & management review. Adv. Onl. Public, 1–36. 10.1080/15309576.2025.2493889

• FarayolaM. M.TalI.SaberT.ConnollyR.BendechacheM. (2025). A fairness-focused approach to recidivism prediction: implications for accuracy, trust, and equity. AI. Soc., 10.1007/s00146-025-02452-1

• FloridiL.CowlsJ.BeltramettiM.ChatilaR.ChazerandP.DignumV.et al (2018). AI4People—an Ethical framework for a good AI society. Minds Mach.28 (4), 689–707. 10.1007/s11023-018-9482-5

• GaoL.Luna-ReyesL. F. (2025). “Implications of adopting AI-based applications in local governments: a systematic literature review from a public value perspective,” in In the Art of Digital Governance, Public Administration, Governance and Globalization, eds. KimY.AhnM. J (Cham: Springer), 363–88. 10.1007/978-3-032-00514-4_18

• GhoshA.SainiA.BaradH. (2025). Artificial intelligence in governance: recent trends, risks, challenges, innovative frameworks and future directions. AI. Soc.40, 5685–5707. 10.1007/s00146-025-02312-y

• Hannah-MoffatK. (2019). Algorithmic risk governance: big data analytics, race and information activism in criminal justice debates. Theoret. Crim.23 (4), 453–470. 10.1177/1362480618763582

• HassijaV.ChamolaV.MahapatraA.GuptaD.SikdarB. (2024). Interpreting black-box models: a review on explainable artificial intelligence. Cognit. Comput.16, 45–74. 10.1007/s12559-023-10179-8

• IkramM.MehmoodW.SaleemS. M. U.HanefarS. B. M. (2026). Exploring the nexus of governance and AI ethics: using systematic literature review for future direction. Data Inf. Manag.10 (3), 100122. 10.1016/j.dim.2026.100122

• KheiriM. A.QashtaN.AljaradatD. I. (2025). Authenticity of using artificial intelligence systems in proving electronic evidence. Pak. J. Life Soc. Sci.23 (1), 37–52. 10.57239/PJLSS-2025-23.1.004

• KoshevaliskaO. (2026). Artificial intelligence in criminal justice: predictive tools, evidentiary challenges and human rights implications. Nauka, Bezbednost, Policija31 (1), 1–19. 10.5937/nabepo31-62922

• LiangG. (2026). Governing the amorphous: aI, social justice, and the challenge of future-proof regulation. Soc. Sci. Humanit. Open13, 102467. 10.1016/j.ssaho.2026.102467

• MukhamadievaG. N.ZhanibekovA. K.ApsimetN. M.AlimkulovY. T. (2025). Integration of artificial intelligence into criminal procedure law and practice in Kazakhstan. Laws.14 (6), 98. 10.3390/laws14060098

• PapagiannidisE.MikalefP.ConboyK. (2025). Responsible artificial intelligence governance: a review and research framework. J. Strat. Inf. Syst.34 (2), 101885. 10.1016/j.jsis.2024.101885

• RajendraJ. B.ThuraisingamA. S. (2025). The role of explainability and human intervention in AI decisions: jurisdictional and regulatory aspects. Inf. Commun. Techol.. Law, 1–32. 10.1080/13600834.2025.2537514

• RoblesP.MallinsonD. J. (2025). Artificial intelligence technology, public trust, and effective governance. Rev. Pol. Res.42, 11–28. 10.1111/ropr.12555

• SierockaH. (2025). Legal and ethical issues related to the use of artificial intelligence in the field of justice. Białostockie Studia Prawnicze30 (4), 177–195. 10.15290/bsp.2025.30.04.11

• SodeK. (2026). The interpreter’s trap: how explainable AI launders uncertainty into justification—a socio-legal case study of COMPAS risk assessment. AI. Soc., 10.1007/s00146-026-02885-2

• TriguiH.GuerfaliF.Harigua-SouiaiE.QasrawiR.AtriC.SokhnE. S.et al (2024). Exploring AI governance in the Middle East and north Africa (MENA) region: gaps, efforts, and initiatives. Data Pol.6, e83. 10.1017/dap.2024.85

• UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence. Paris: United Nations Educational, Scientific and Cultural Organization. Available online at:https://www.unesco.org/en/articles/recommendation-ethics-artificial-intelligence (Accessed April 23, 2026).

• VatamanuA. F.TofanM. (2025). Integrating artificial intelligence into public administration: challenges and vulnerabilities. Adm. Sci.15 (4), 149. 10.3390/admsci15040149

• VidakiA. N.PapakonstantinouV. (2025). Democratic legitimacy of AI in judicial decision-making. AI. Soc.40, 6025–6035. 10.1007/s00146-025-02411-w

• WidłakT. (2026). Techno-Legal virtues in hybrid adjudication: judicial excellence in AI-augmented courts. Minds Mach.36, 3. 10.1007/s11023-025-09756-7

• ZajkoM. (2021). Conservative AI and social inequality: conceptualizing alternatives to bias through social theory. AI. Soc.36, 1047–1056. 10.1007/s00146-021-01153-9