Public participation in artificial intelligence (AI) development by public administrations is increasingly proposed as a potential silver bullet for mitigating the risks associated with the use of AI in the public sector. This paper examines whether public participation can serve as a viable solution from a doctrinal and practical perspective by analyzing the availability of legal avenues for natural and legal persons to participate in AI development procedures by the EU and member states’ public entities. While the idea is appealing, there are significant obstacles. Notably, EU law currently lacks specific legal frameworks to enable public involvement in the design and development of AI systems by public entities. Neither traditional European administrative law nor the new Artificial Intelligence Act provides effective legal avenues for involving the broader public in AI development in the public sector, posing challenges to implementing meaningful participation. Furthermore, drawing an analogy from environmental law, where public participation often ends up being formalistic and ineffective, suggests that similar outcomes might arise in the context of Public-Sector AI development. Therefore, despite the theoretical advantages of involving the public, this paper argues that, without robust legal structures and meaningful procedural designs, public participation in Public-Sector AI development may ultimately fall short of achieving its promises, proving to be a dead end.

Artificial intelligence (AI) systems already govern – they rank speech, allocate attention, filter applicants and triage claims. Technical improvements in safety and alignment do not determine whether publics will accept that governance. Legitimacy, on the sociological view used here, turns on recognition: whether power appears rule-bound, justified and situated in institutions that speak for those it governs. Social media demonstrates the problem: distribution improved even as publics questioned who sets the rules, producing recurrent conflicts grounded in missing authorization. As AI embeds upstream of more domains, the risk is repeating that crisis at larger scale – systems that perform yet lack legitimacy, inviting resistance and regulatory backlash. This Article makes legitimacy an explicit regulatory aim for AI. It identifies three persistent deficits: opacity blocks audience-understandable reasons; private actors wield public-facing authority without recognized authorization; and administrative automation strains participation, reason-giving and review. Legal tools can structure recognition. Thin legality supplies visible forms – publicity, stability, consistency – that signal non-arbitrariness. Thick legality adds public authorship, audience-facing reasons and contestation. This Article distills this into three principles: Integration seats AI rule-setting in recognized venues; Familiarity presents rules in locally credible forms; and Contestation guarantees a credible second look. Together, these convert performance into justification and justification into authority publics recognize as rightful.

While artificial intelligence (AI) offers promise as a tool for efficiency and access to justice, in public administration it also threatens to reproduce the perversity in Franz Kafka’s parable “Before the Law”: systems that mimic legality while rendering the law opaque and unreachable. I develop the concept of “Digital Kafkaism” to describe this pathology, and argue that a crucial surface of contest lies in the design of human–computer interactions that mediate administrative decisions. Drawing on administrative law’s core values (such as reason-giving, accountability and contestability), I argue that these can be preserved within socio-technical systems if we treat Responsible AI (RAI) not as a matter of stating abstract principles or technical fixes, but as a design brief. Lawyers, I argue, have a distinctive contribution to make in this interdisciplinary project: to ensure that the gateway to the law remains genuinely open by embedding public law commitments in the architectures of digital administration. To this end, I propose a programmatic set of design-facing requirements to translate public-law values into operational constraints for AI-mediated administration.

This article explores attempts to regulate the use of “generative” artificial intelligence (GenAI) by the judicial arm of government in jurisdictions in Australasia. There are significant concerns about the use of GenAI-produced materials in courts, including wasting the courts’ resources, fairness concerns and undermining public trust in legal processes. Courts such as those in Singapore, New Zealand and the Australian states of Victoria, Queensland, New South Wales, South Australia and Western Australia have issued guidance or practice notes seeking to direct or limit the use of GenAI tools. Other courts have not issued guidance, but drawn on a range of regulatory strategies to respond to the use of GenAI in proceedings. In this article, we focus on four Australasian jurisdictions: Australia, Hong Kong, New Zealand and Singapore. Responsive regulation invokes the concept of a pyramid of regulatory mechanisms, moving from lower-level options aimed at education and motivating compliance through to punitive measures designed to deter and even to punish. The variety of regulatory strategies courts have adopted includes prohibition, disclosure and awareness-raising. Like the genie, GenAI will not now go back into the bottle as both lawyers and self-represented parties continue to make use of it. We therefore address the future feasibility of current regulatory strategies in this context, and the importance of continuing to refine the regulatory approach.

Artificial intelligence (AI) is increasingly used in public administration, yet its opacity raises concerns about arbitrary exercises of state power. This article argues that alongside regulatory frameworks, the Rule of Law supplies an independent basis for limiting state deployment of AI. Drawing on widely accepted ideas associated with the Rule of Law – predictability (decisions must guide conduct ex ante) and procedural clarity (rationales must be knowable and contestable ex post) – the article shows how these standards constrain government use of AI. It examines why black box systems frustrate legal guidance and contestability, and why explanations are often the minimally sufficient remedy. A welfare-benefits hypothetical illustrates the stakes of opacity versus explainability. The article concludes that Rule of Law limits offer a distinct and operationalizable basis for determining when state deployment of AI contravenes fundamental commitments, demonstrating that not all government uses of AI should be permitted.

This article investigates how to translate abstract responsible artificial intelligence (AI) principles into practical constraints for public-sector AI systems. Focusing on due process, transparency, fairness, reason-giving and proportionality, the paper uses administrative and constitutional law doctrines to scrutinize AI deployments in government. The purpose of this comparative analysis is to evaluate how the divergent frameworks of the United States (US), the European Union (EU) and Singapore fulfill responsible AI principles. The paper contrasts the EU’s unified hard-law model with the fragmented preemption approach of the US and the facilitative soft-law guidelines of Singapore. Furthermore, the analysis explores how legal norms can be integrated throughout the AI lifecycle, addressing the inherent tensions between technical performance and legal compliance. Illustrative case studies – including the Dutch childcare-benefits algorithm, the UK exam-grading fiasco, the US Internal Revenue Service audit controversies and Estonia’s successful Kratt AI implementation – highlight the severe consequences of algorithmic opacity and the benefits of proactive governance. Drawing on these insights, the paper argues that multidisciplinary governance is essential. The article’s original scholarly contribution lies in systematically mapping administrative and constitutional law doctrines onto AI requirements across the AI lifecycle, proposing a concrete techno-legal architecture that ensures automated public decisions comply with enduring legal norms.