This article discusses the rise in the use of Gen artificial intelligence (Gen AI) for the production of mandatory corporate reporting, particularly narrative and ESG (Environmental, Social or Governance) reports. The capabilities of Gen AI can potentially deliver many benefits, but firms are exposed to legal and regulatory risks in connection with Gen AI adoption. This article discusses how firms may address these risks, but more importantly, these risks should not be appreciated only at the firm-level but at a broader industry, market and systemic level. When viewed through the lens of the reporting chain, which is the universe of recipient entities that use such corporate reporting, including numerous financial intermediary entities, regulators and finfluencers, these risks take on new implications that require regulatory and supervisory efforts for their oversight and mitigation. The article makes specific proposals for securities and financial regulators in particular, against a broader context of the more general, cross-cutting nature of AI systems regulation and governance.

The establishment of artificial intelligence regulatory sandboxes (AIRSs) poses both policy and technical challenges, especially in how to reconcile support for innovation with regulatory oversight. AIRSs are based on dynamic regulatory feedback mechanisms that allow for a deeper examination of legal norms with a view to their future evolution. These structures facilitate engagement between regulators and innovators, enabling business learning and regulatory adaptation. However, their proliferation across the European Union under the Artificial Intelligence Act (AI Act) may raise issues of coordination between competent authorities, cross-border regulatory alignment and consistency with overlapping (sectoral) rules. In view of these potential complexities, this paper makes two distinct recommendations. First, AIRSs would benefit from cross-border cooperation – efforts should therefore be made to pursue the establishment of joint AIRSs among different Member States in order to reduce regulatory fragmentation, lower the risk of forum shopping, and optimise administrative resources. Second, integrating AI and cybersecurity compliance within the same sandbox environment would be beneficial in terms of providing clearer and more structured compliance pathways. A well-designed regulatory sandbox regime would make regulation more effective, encourage responsible AI development and secure Europe’s leadership in digital regulation.

Regulatory sandboxes emerged as prominent tools for fostering financial innovation, particularly in jurisdictions where regulatory rigidity may hinder technological advancements. This article examines the Brazilian Central Bank’s experience with regulatory sandboxes between 2021 and 2024, analyzing their role in balancing market modernization with institutional oversight. While often presented as a panacea for regulatory challenges in emerging markets, the effectiveness of these frameworks remains a subject of debate. By situating its approach within broader discussions on regulatory experimentalism, this study explores whether sandboxes serve merely as deregulatory mechanisms or can function as state-led instruments for fostering financial innovation. Drawing from the concept of the entrepreneurial state, the article investigates how specialized regulatory agencies can collaborate with private actors to drive technological development in financial markets. The findings suggest that, while the BCB’s regulatory sandbox challenges the traditional dichotomy between regulation and innovation, offering a model where legal oversight and market experimentation coexist, it also raises concerns about regulatory capture, market concentration, and the potential erosion of consumer protections. Although sandboxes can stimulate innovation, their long-term success depends on mitigating these risks and ensuring that financial experimentation does not come at the expense of broader public interest objectives.

Regulatory sandboxes for Artificial Intelligence (AI) are designed to address challenges of rapid technological change. AI innovations create an acute need for learning about what regulation is suitable for enabling innovation while dealing with technological risks. This article argues that regulatory sandboxes should be analyzed primarily as mechanisms for enhancing policymakers’ understanding of technologies such as AI, rather than solely as spaces for experimentation that promote innovation. It discusses the role of regulatory sandboxes in facilitating policy learning that can complement the long-term learning processes of the traditional policy cycle. Six case studies serve to illustrate sandbox elements for enabling collaborative experiential learning in contexts in which the absence of AI regulation makes accelerated policy learning particularly valuable. Looking at the design and governance of regulatory sandboxes from Brazil, Colombia, Mauritius, Mexico, Rwanda, and Thailand, learning elements related to the technology and consequences for closing legal lags emerge as critical components.

The article analyzes whether regulatory sandboxes are a legally intended and useful instrument to create pathways for law enforcement agencies using cutting-edge AI systems in compliance with fundamental rights. It takes a deeper look at the provisions of the EU AI Act for regulatory sandboxes and testing under real-world conditions in contexts relevant to law enforcement. Using a legislative process in Germany as a case study, the article shows the imminent tension of innovative AI systems between modern investigation tools in the digital age, as well as potential pathways towards disproportionate surveillance of citizens. The author suggests taking the extra mile through regulatory sandboxes instead of crafting legal foundations quickly under the influence of current political pressure: On the long run this can save tax money, avoid dependencies to tech companies and pave a European way of AI systems that respect digital human rights in an appropriate manner.

How can existing experiences of regulatory experimentation inform AI sandbox design in Europe? This paper explores the ‘responsible AI’ sandbox of the Norwegian Data Protection Agency (DPA), a GDPR-oriented regulatory experiment created in 2020 with four projects per annum. Through an interpretive policy analysis of documents (exit reports and workshop transcripts) and semi-structured interviews with officials, we explore how the Norwegian DPA approached its mandate of ‘helping with responsible innovation’, where it identified role conflicts, and what scope conditions and challenges it perceived around sandbox work. Sandboxing represented a ‘new way of working’ for the regulatory authority: in an idea-based intervention mode, the DPA moves from rule-based interventions as a watchdog to becoming a dialogue-oriented partner in solution-finding, a concretiser of ambiguous GDPR rules, and a keen learner from sectoral and technical experts. Critical engagement with our data suggests that sandbox design should not be reduced to technical and procedural questions. It requires regulators’ critical reflexivity on their ambivalent role and power relations in the regulatory experiment: how to strategically select relevant projects and issues, how to navigate budgetary constraints and the lack of follow-ups, and how sandboxing affects more interventionist regulatory duties.

Modern Elections can be conceived as a socio-technical system, as the electoral process in many ways relies on technological solutions: voter information, identification and registration, collecting, verifying and counting the votes – in some countries these steps are conducted by using innovative technologies. But how do those devices and processes actually become part of the official legislation and can finally deployed during this sensitive and important democratic procedure? Over time, the State of California has developed a robust regulatory ecosystem for integrating innovative technology into the electoral process and is also able to change and modernize its rules and regulations. Although technologies currently used are more static, hardware-based and usually do not include algorithmic systems, the overall structure of the process may also function as a blueprint for regulating more dynamic algorithm-based or even AI-based technologies.

Against the backdrop of the European Union (EU)’s objective that artificial intelligence (AI) regulatory sandboxes under the EU AI Act should both foster innovation and assist with compliance, considerable challenges are identified for Member States around capacity-building and the design and operationalisation of regulatory sandboxes. The article charts Member States’ early-stage preparations for implementing AI regulatory sandboxes and contends that there is a risk that differing approaches being taken by individual national sandboxes could lead to fragmentation and jeopardise a uniform interpretation of the AI Act in practice. This could also motivate innovators to play sandbox arbitrage. With sandbox participation being voluntary, the possibility that AI regulatory sandboxes may prove unattractive to innovators on their compliance journey is also explored. Confidentiality concerns, the inability to relax legal rules during the sandbox and the inability of AI regulatory sandboxes to deliver a presumption of conformity with the AI Act are identified as pertinent concerns for innovators contemplating applying to AI regulatory sandboxes as compared with other direct pre-market compliance routes provided to them in the form of application of harmonised standards and conformity assessment procedures.

This paper provides a comprehensive case study of Israel’s emerging regulatory sandbox frameworks for AI applications. Drawing from policy white papers, legislative materials, and implemented frameworks, it documents how Israel is adopting sector-specific regulatory sandboxes as experimental mechanisms to balance innovation with public interest concerns. The paper maps the growing appeal of regulatory sandboxes across Israeli AI policy documents since 2018 and provides a detailed analysis of three key implementations: autonomous vehicles, fintech, and general legislative guidelines. It examines Israel’s pioneering AV regulatory sandbox, which grants specific exemptions from traffic regulations (including steering wheel control requirements and international standards compliance) while maintaining safety oversight through control centers and incident reporting. The paper analyzes the proposed fintech sandbox’s multi-regulator committee structure and dual-track system for licensing and oversight relief. It also evaluates the recently published general regulatory sandbox legislative guidelines that establish a technology-neutral four-phase lifecycle framework. The analysis reveals Israel’s preference for adapting existing regulatory structures rather than creating comprehensive AI-specific governance mechanisms. While this sector-specific approach facilitates targeted innovation, it may inadequately address broader AI governance concerns such as algorithmic accountability and human oversight. The paper critically examines institutional design challenges, including inter-agency coordination difficulties in the fintech framework and the tension between innovation facilitation and consumer protection. It concludes that Israel’s approach represents a pragmatic response to technological uncertainty, though questions remain about implementation effectiveness and the adequacy of sector-specific frameworks for comprehensive AI governance. The study contributes to comparative regulatory literature by documenting an understudied jurisdiction’s approach to regulatory experimentation.

The Artificial Intelligence (AI) Act introduces AI regulatory sandboxes and testing in real-world conditions as experimentation tools to support innovation in AI development. While regulatory sandboxes have been widely adopted across various sectors, their precise function and scope under the AI Act remain complex and multifaceted. This paper examines the legal framework governing AI regulatory sandboxes, analysing the degree of regulatory flexibility they provide and how they interact with other experimentation tools established by the AI Act. In particular, the study assesses whether these instruments enable AI systems to be tested under reduced regulatory constraints or whether they primarily serve as forums for regulatory dialogue. It then explores the differences between AI regulatory sandboxes and the testing in real-world conditions mechanism, highlighting how sandboxes focus on legal guidance, while testing in real-world conditions aims to remove legal barriers and facilitate market entry. The analysis also identifies key interpretative challenges, including the interplay between Article 57 (AI regulatory sandboxes) and Article 60 (testing in real-world conditions), the compliance requirements for each mechanism, and the role of regulatory authorities. The paper concludes that further clarification from the European Commission is necessary to address inconsistencies in the AI Act’s provisions, ensuring effective implementation of AI regulatory sandboxes and testing in real-world conditions.

Manipulation and deception were not born with AI: online architecture of choice can be harmful when it contains dark patterns or deceptive designs. These techniques deceive or manipulate users through interfaces that have the substantial effect of subverting or altering users’ agency, decision-making, or choice as part of their online activities. But AI has the potential to further enhance this manipulation increase its sophistication and scale. This article presents the principle of ‘Fairness by Design’ as a potential solution as well as a set of interface prototypes inspired by it and developed within Amurabi’s R&D Lab. These solution prototypes are called ‘Fair Patterns’. Fair patterns make it possible to implement the principles of transparency, trust, and autonomy by providing the right level of information at the right moment in the user journey, in clear language and without cognitive overload.

This paper begins with a brief technical explanation of generative AI and how only a small subset of entities – the landlords of creativity – have access to the computational resources and expertise needed to create foundation models that enable audio deepfakes. It then examines how regulatory regimes in America, the European Union (EU), and China address the misuse of generative AI. Although each framework seeks to regulate generative AI in different ways, the paper argues that none effectively assigns liability to the landlords of creativity. Finally, the paper proposes holding these landlords responsible for their renters’ malicious usage. This proposal not only is technically feasible but also is conceptually aligned with established legal doctrines in the American, EU, and Chinese frameworks.

In this article, we investigate the relationship between human oversight and fairness within the evolving framework of EU AI regulatory governance. We address two core research questions: (1) How are human oversight and fairness related? and (2) To what extent does the AI Act establish a framework for human oversight that effectively supports the implementation of the various dimensions of fairness? Based on a review of interdisciplinary literature, the article identifies three normative claims linking human oversight to fairness: first, that human oversight can help mitigate bias and error in AI systems; second, that it can function as a mechanism of accountability by assigning oversight to natural persons where AI systems lack legal liability; and third, that it can introduce human empathy and contextual sensitivity into decision-making processes, enabling a substantive notion of fairness that takes into account individual circumstances. A critical analysis of the AI Act reveals that while these normative aspirations are acknowledged, the Act only partially operationalises them, leaving several aspects of fairness insufficiently supported.

Fairness in service robotics is a complex and multidimensional concept shaped by legal, social and technical considerations. As service robots increasingly operate in personal and professional domains, questions of fairness – ranging from legal certainty and anti-discrimination to user protection and algorithmic transparency – require systematic and interdisciplinary engagement. This paper develops a working definition of fairness tailored to the domain of service robotics based on a doctrinal analysis of how fairness is understood across different fields. It identifies four key dimensions essential to fair service robotics: (i) furthering legal certainty, (ii) preventing bias and discrimination, (iii) protecting users from exploitation and (iv) ensuring transparency and accountability. The paper explores how developers, policymakers and researchers can contribute to these goals. While fairness may resist universal definition, articulating its core components offers a foundation for guiding more equitable and trustworthy human–robot interactions.

An emerging field of AI, namely Fair Machine Learning (ML), aims to quantify different types of bias (also known as unfairness) exhibited in the predictions of ML algorithms, and to design new algorithms to mitigate them. Often, the definitions of bias used in the literature are observational, i.e. they use the input and output of a pre-trained algorithm to quantify a bias under concern. In reality, these definitions are often conflicting in nature and can only be deployed if either the ground truth is known or only in retrospect after deploying the algorithm. Thus, there is a gap between what we want Fair ML to achieve and what it does in a dynamic social environment. Hence, we propose an alternative dynamic mechanism, “Fair Game”, to assure fairness in the predictions of an ML algorithm and to adapt its predictions as the society interacts with the algorithm over time. “Fair Game” puts together an Auditor and a Debiasing algorithm in a loop around an ML algorithm. The “Fair Game” puts these two components in a loop by leveraging Reinforcement Learning (RL). RL algorithms interact with an environment to take decisions, which yields new observations (also known as data/feedback) from the environment and in turn, adapts future decisions. RL is already used in algorithms with pre-fixed long-term fairness goals. “Fair Game” provides a unique framework where the fairness goals can be adapted over time by only modifying the auditor and the different biases it quantifies. Thus, “Fair Game” aims to simulate the evolution of ethical and legal frameworks in the society by creating an auditor which sends feedback to a debiasing algorithm deployed around an ML system. This allows us to develop a flexible and adaptive-over-time framework to build Fair ML systems pre- and post-deployment.

As judges increasingly rely on algorithmic and AI systems in their decision-making process, concerns have been raised about their impact on the judicial duty to state reasons – a duty that is essential to the right to a fair trial and the rule of law. This paper examines whether existing legal frameworks adequately ensure this duty and its underlying normative goals in the age of automation and explores how best to safeguard this duty. Central to the analysis is the question of whether the duty to state reasons itself should be redefined and impose stricter reasoning requirements on judges using AI, requiring them to articulate both ‘pragmatic’ and technical explanations of the systems’ role in their decisions. While this approach may enhance the duty’s normative goals of legitimacy, transparency and accountability, the paper also considers the potential drawbacks of imposing such a more robust reasoning obligation in AI-assisted judicial processes.

Amid the rapidly evolving landscape of artificial intelligence (AI) regulation, a significant concern has emerged regarding the predominant focus on preemptive measures aimed at preventing or mitigating potential AI-related harms. While these preemptive measures are undeniably important, they must be complemented by effective redress mechanisms to address and remedy materialized harms. This paper highlights the crucial role of redress in empowering individuals to challenge and rectify the adverse effects of AI systems, emphasizing that access to redress is, in fact, access to justice. We critically evaluate whether current AI governance sufficiently address the need for remedies for AI-related harms, arguing that they fall short in protecting individuals’ rights. To address this gap, we outline four key steps in the redress process: (1) initiating the redress process, (2) determining appropriate avenues for redress, (3) collecting evidence to support claims and (4) receiving and responding to decisions. Each step is explored in detail, presenting distinct challenges and requirements, illustrated with real-world examples. Our findings underscore the urgent need to integrate robust redress pathways into AI governance frameworks to safeguard individual rights as AI technologies become increasingly embedded in society.