The Algorithmic State: Proxy Governance, Synthetic Actors, and the Future of Public Administration
Introduction: The Advent of Algorithmic Statecraft
Modern statecraft is undergoing a profound structural transformation that fundamentally alters the mechanisms of public administration and political representation. For decades, the digitalization of government services—often termed "e-government"—focused primarily on the automation of routine analog processes, transitioning paper-based workflows into digital databases to improve citizen accessibility and bureaucratic efficiency. However, the contemporary integration of artificial intelligence (AI) represents a categorical leap from passive digitalization to proactive algorithmic statecraft. Artificial intelligence systems are no longer deployed merely as analytical tools utilized by civil servants; they are increasingly positioned as quasi-autonomous agents capable of exerting executive, judicial, and administrative authority. This evolution has given rise to the phenomenon of "proxy governance," wherein sovereign states delegate complex decision-making, resource allocation, and even political representation to synthetic algorithmic systems. The integration of artificial intelligence into the machinery of the state is accelerating globally, though its implementation remains highly asymmetrical. Recent comprehensive analyses by the Organisation for Economic Co-operation and Development (OECD) indicate that 57% of government AI use cases focus on automating, streamlining, and tailoring public services, while 45% are designed to enhance decision-making, sense-making, and long-term forecasting1. Furthermore, 30% of state AI initiatives aim to improve accountability and anomaly detection, such as identifying fraud within tax administration or public welfare distribution1. Only 4% of analyzed cases currently allow external actors to leverage government AI to achieve their own independent goals, exemplified by initiatives like Greece's DidaktorikaAI, which provides an AI-powered library of 50,000 publications1. Despite these advancements, the adoption of AI by governments has historically lagged behind the private sector. This delay is attributed to pervasive skill gaps within the civil service, the persistence of legacy IT systems, fragmented data silos, tight budgetary constraints, and the stricter regulatory requirements necessary to ensure privacy, transparency, and democratic representation1. Compounding these challenges is a widespread lack of impact measurement frameworks, which prevents many government AI initiatives from scaling beyond the pilot phase by failing to demonstrate a clear return on investment2. Nevertheless, the frontier of governmental AI has rapidly expanded into the core functions of state power. Governments are moving beyond high-volume transaction areas like basic civic participation and are now deploying AI to optimize urban infrastructure, triage complex constitutional legal petitions, manage geopolitical strategic dependencies, and, in unprecedented administrative experiments, act as synthetic political candidates and cabinet-level ministers. This transition fundamentally alters the phenomenology of the state. Traditional Weberian bureaucracies derive their legitimacy from human discretion, legal accountability, and a hierarchical chain of command that makes the complex world legible for state intervention3. In contrast, the emerging "dataist state" relies on probabilistic inference, machine learning models, and dynamic predictive analytics3. While this promises unprecedented efficiency and the theoretical eradication of human error, it simultaneously introduces profound vulnerabilities. Algorithmic governance threatens to obscure accountability, embed historical biases into mathematical models, and shift political power away from democratically elected representatives toward the technocratic architects of the AI infrastructure. As artificial intelligence is increasingly tasked with running countries by proxy, the central challenge for contemporary public administration is reconciling the inherent opacity of algorithmic decision-making with the democratic imperatives of transparency, accountability, and the rule of law.
Theoretical Frameworks: Institutionalizing Proxy Responsibility
To comprehend the implications of artificial intelligence as a political and administrative actor, it is necessary to establish the theoretical frameworks that govern machine-mediated authority. The integration of AI into high-stakes state functions—ranging from public procurement to judicial triage and military resort-to-force decisions—creates what legal and political scholars identify as a severe socio-technical "responsibility gap"5.
The Socio-Technical Responsibility Gap and Synthetic Agents
When human bureaucrats or military commanders make decisions, they are bound by legal liability, ethical norms, and political accountability. If a human actor makes a catastrophic error, the legal and administrative systems possess established mechanisms to attribute blame and enforce consequences. However, AI systems, acting as "synthetic agents," do not qualify as moral or legal agents5. Due to the inherent opacity of deep neural networks and the "problem of many hands"—a scenario where countless developers, data scientists, procurement officers, and policy integrators contribute to a single algorithmic system—it becomes nearly impossible to attribute direct moral or legal culpability when an algorithmic decision causes harm5. To bridge this persistent gap, contemporary governance theorists propose the institutionalization of "proxy responsibility." Proxy responsibility dictates that specific human actors, or designated organizational oversight bodies, must formally assume legal and moral responsibility for the decisions generated by a synthetic agent that cannot hold responsibility itself5. Unlike distributed responsibility, which can dilute accountability across an organization, proxy responsibility focuses on establishing the strict preconditions necessary for attributing responsibility in the first place. For proxy responsibility to function effectively, the human overseers must possess comprehensive knowledge of the system's operational parameters, the unencumbered autonomy to override algorithmic recommendations, and the intentionality to guide the system's outcomes5. Establishing these preconditions is exceptionally difficult in modern public administration. Developers of AI systems are generally unaccustomed to considering the complex ethical, political, and constitutional constraints of statecraft, while politicians and military leaders often lack the technical literacy required to understand how these complex models function5. If these preconditions are not met, the concept of a "human-in-the-loop" devolves into a dangerous administrative illusion. Instead of providing meaningful oversight, human operators succumb to "automation bias," defaulting to a process of mere "rubber-stamping" where they blindly defer to machine outputs out of convenience or an unwarranted trust in algorithmic infallibility6.
The AI Oversight Body and the Prevention of Agency Laundering
To operationalize proxy responsibility, scholars advocate for the integration of specialized AI oversight bodies directly into the organizational decision-making structures of the state5. Such bodies must be situated at the nexus of the political, military, legal, and economic systems, serving as interdisciplinary centers of expertise5. The composition of an AI oversight body must mirror the actors involved in the broader decision-making process, ensuring that technical experts, legal scholars, and political leaders can translate technical data into actionable political language5. The primary function of this oversight body is to prevent the phenomenon of "agency laundering" or scapegoating. When AI is integrated into the state apparatus, individual or collective human actors may attempt to use the algorithmic system as a scapegoat, "passing the buck" and feeling relieved of their own responsibility when a policy fails or a controversial decision is made5. By explicitly attributing proxy responsibility, the oversight body ensures that if an AI system incorrectly assesses a geopolitical threat and leads politicians to make a flawed resort-to-force decision, a case-specific assignment of responsibility remains possible. Accountability can then be traced back to the developers who ignored ethical constraints, the integrators who failed to communicate technical limitations, or the politicians who abdicated their decision-making authority5.
| Bureaucratic Responsibility vs. Proxy Algorithmic Responsibility | | :---- | | Traditional Bureaucratic Model: Direct attribution of responsibility to a human agent based on hierarchical authority, legal mandates, and discretionary intent3. | | The Responsibility Gap: The inability to assign moral or legal blame to a synthetic agent (AI) due to algorithmic opacity and the "problem of many hands"5. | | Automation Bias Risk: Human operators devolving into "rubber-stampers" who blindly approve AI outputs without exercising true scrutiny or understanding the underlying logic6. | | Proxy Responsibility Framework: Human actors or specialized oversight bodies formally absorbing the legal and moral liability for decisions generated by an AI system5. | | Prevention of Agency Laundering: Ensuring that politicians and administrators cannot use an algorithm as a scapegoat to deflect accountability for failed policies or harmful actions5. |
The Architecture of Algorithmic Statecraft and Surveillance Federalism
Beyond the mechanics of accountability, the integration of artificial intelligence into government fundamentally reconfigures the distribution of power, giving rise to new political science paradigms such as "algorithmic statecraft" and "surveillance federalism." AI must be conceptualized not merely as a neutral efficiency technology, but as a power-bearing political infrastructure that institutionalizes a regime of probabilistic inference, classification, and attention allocation4.
The Upstream Migration of Administrative Discretion
In traditional public administration, discretion is exercised by front-line civil servants and agency directors who interpret policies based on contextual human judgment. Under algorithmic governance, this discretion undergoes a radical transformation. Authority migrates "upstream" into the initial phases of objective-setting, proxy construction, threshold calibration, and lifecycle updating of the AI models4. When a government deploys an AI system to allocate public welfare or determine tax audit targets, the political choices are embedded deeply within the code. The selection of training data, the weighting of variables, and the definition of the algorithm's objective function represent profound political decisions made by data scientists and procurement officers, often insulated from public scrutiny or democratic contestation4. This upstream migration generates severe contestability deficits, administrative burdens, and accountability gaps, which are frequently intensified by procurement lock-in and the proprietary enclosure of algorithms by private technology vendors4.
Surveillance Federalism and the Erosion of Local Autonomy
The deployment of centralized algorithmic statecraft also alters the delicate balance of power between national governments and local municipalities, resulting in what researchers term "surveillance federalism"8. Using qualitative comparative designs analyzing nations like India, Singapore, Indonesia, and the Philippines, scholars demonstrate how centrally controlled digital architectures radically transform the federal bargain8. Systems such as biometric identity regimes and centralized algorithmic governance platforms purportedly improve the efficiency of public service provision. However, they simultaneously erode the sovereignty of sub-national governments. The consent-based shared sovereignty that characterizes strong democratic federalism is replaced by technologically mediated, centralized controls8. Surveillance federalism operates through three interlocking processes: the centralisation of data at the expense of local democratic institutions; the emergence of technocratic public administration that replaces deliberative politics with computational rationality; and the establishment of an algorithmic paradigm that reduces citizens to mere data inputs8.
Algorithmic Diplomacy and Geopolitical Asymmetry
The theoretical implications of algorithmic statecraft extend into international relations, reshaping the practice of diplomacy. Classical International Relations (IR) paradigms—such as realism, liberalism, and constructivism—have historically treated technology as a passive instrument of statecraft10. However, AI systems act as influential factors that actively shape the decision-making environment, introducing the concept of "algorithmic diplomacy"10. When AI systems are embedded in the core processes of diplomacy and security, they introduce new logics and priorities that may diverge from the interests of their human deployers. The logic of an algorithm might systematically privilege certain outcomes and occlude others before any human diplomat has made a conscious choice10. This dynamic manifests in three critical areas: algorithmic bias within international organizations that perpetuates structural inequalities between the Global North and South; the incorporation of AI into nuclear command and control systems, which radically compresses crisis management timelines and alters the logic of Mutually Assured Destruction (MAD); and the emergence of computational power imbalances in global trade and climate negotiations10. Furthermore, this geopolitical landscape has given rise to "Technological Swing States" (TSS)—middle powers such as South Korea, Singapore, and India that possess both technological capacity and strategic flexibility11. These nations reconceptualize AI opacity not as a technical deficit, but as a strategic resource. By leveraging the structural opacity of algorithms, these states navigate U.S.-China techno-competition through strategies of delay, selective alignment, and normative intermediation, effectively converting technical constraints into significant diplomatic leverage11.
Municipal Automation and Platform Urbanism: The "City Brain" Paradigm
The most comprehensive and tangible realization of algorithmic statecraft currently operating at the municipal level is found in the deployment of urban artificial intelligence. The "City Brain" architectures, pioneered in the People's Republic of China, exemplify how the theoretical constructs of algorithmic governance are applied to physical infrastructure. A modern city is an aggregate of a massive volume of heterogeneous data, and the primary challenge for municipal public administration is extracting meaningful, real-time value from this data to govern urban populations effectively12.
Global Cognition and Algorithmic Urban Control
The paradigm of the City Brain was established in 2016 when the Chinese tech giant Alibaba Cloud developed the "ET City Brain" and deployed it in the city of Hangzhou, in the Zhejiang province13. The City Brain is defined as an end-to-end autonomous governance system that utilizes vast arrays of sensors, surveillance cameras, and digital infrastructure to ingest ultra-large-scale, multi-source data streams12. The system operates on three foundational metrics that distinguish it from standard smart city initiatives. First, it achieves "global cognition" by processing multi-source data at a scale and speed that human administrators cannot comprehend in real time. Second, it utilizes deep neural networks and machine learning to uncover complex hidden rules and urban dynamics that humans have not discovered. Finally, it possesses the capacity to autonomously formulate and execute globally optimal intervention strategies12. In Hangzhou, the City Brain was initially tasked exclusively with the management of urban traffic. By analyzing live video streams from hundreds of CCTV cameras and intersecting that intelligence with urban maps and real-time weather forecasts, the AI developed a comprehensive situational awareness of the city's mobility13. The algorithms subsequently assumed direct control over more than 1,000 road signals14. The results of this algorithmic intervention were profound: the system achieved a 92% accuracy rate in incident identification, reduced traffic congestion by 15%, shortened daily commutes by three minutes, and increased overall travel speeds by 8% to 15%13. Hangzhou, previously ranked as the fifth most-congested city in China, fell to 57th place following the system's launch14. Furthermore, the City Brain autonomously prioritized routing for emergency vehicles, reducing the arrival time of ambulances by half and allowing rescue teams to reach their destinations seven minutes earlier than before14.
Platform Urbanism and the Expansion of State Control
The operational success of the City Brain in traffic management catalyzed its rapid expansion into broader domains of environmental and social governance. The platform evolved to include modules such as the "Environment Brain," which combines geolocation systems with environmental data to anticipate waste production, predict the capacity of urban photovoltaics, and foresee the carbon footprints of private companies13. This expansion illustrates the core, scalable logic of algorithmic statecraft: once a central AI platform achieves situational awareness of a complex urban system, its predictive and autonomous capabilities can be applied to virtually any domain of public administration13. The City Brain model has now been expanded to control aspects of urban governance that include healthcare monitoring, supply chain logistics, and municipal finance13. This phenomenon, described by academics as "platform urbanism," effectively turns the city into a self-managing entity16. However, it also introduces profound shifts in state-society relations. The City Brain paradigm operates on a strictly technocratic rationale that bypasses public deliberation. Decisions regarding resource allocation, infrastructure investment, and citizen mobility are made instantaneously by proprietary algorithms operating within a "black box," fundamentally altering the nature of civic participation8. The export of this model beyond China's borders demonstrates the global appeal of predictive statecraft. The Alibaba City Brain has been exported to over twenty cities globally, including a major deployment in Kuala Lumpur, Malaysia13. This international proliferation signals a growing consensus among governments that the optimal method to manage complex urban populations is through continuous, data-driven algorithmic intervention.
The Agentic State: The United Arab Emirates' Executive AI Integration
While municipal systems like the City Brain focus primarily on physical infrastructure and urban optimization, national governments are beginning to delegate core executive and bureaucratic functions directly to artificial intelligence. The United Arab Emirates (UAE) has aggressively positioned itself at the vanguard of this movement, orchestrating a transition from predictive analytics to fully autonomous "agentic AI" embedded within federal operations.
Mandating the Agentic State and the AI & Data Authority
The UAE has formally initiated a sweeping directive, announced by UAE Prime Minister Sheikh Mohammed bin Rashid Al Maktoum, to have 50% of its federal government operations powered by agentic AI within a strict two-year timeline18. Unlike standard generative AI—which serves merely to produce text, summaries, or analytical reports for human review—agentic AI introduces autonomous systems capable of independently analyzing data, formulating decisions, executing actions, and iterating on those actions in real time without continuous human intervention18. To facilitate this unprecedented restructuring of the state apparatus, the UAE established the Artificial Intelligence and Data Authority. This centralized, cabinet-level body is led by Omar Sultan Al Olama, who was appointed in 2017 as the world's first Minister of State for Artificial Intelligence20. The Authority was created by consolidating the functions of three previously separate entities: The Office of Artificial Intelligence, the Digital Government Sector, and the UAE Data Office20. Its mandate is to unify public data, digital government capabilities, and AI deployments into a single national ecosystem20. The rhetoric emanating from UAE leadership explicitly redefines the relationship between the state and technology. Sheikh Mohammed stated that the goal is to build a government that "runs on data and agentic AI," describing AI no longer as a mere tool, but as an "executive partner" embedded within the core machinery of governance18. The deployment of virtual entities, such as the AI-generated government spokesperson "Zayed"—serving as a "virtual champion" of the UAE Presidential Court's strategic objectives—further underscores the state's commitment to manifesting a heavily AI-driven public identity22.
The 4-Layer Infrastructure Readiness Model and the Hybrid Workforce
The UAE's integration of agentic AI is designed to resolve a persistent bottleneck in public-sector modernization: institutional capacity. Rather than treating automation as a superficial layer atop existing bureaucratic silos, the state is actively executing a plan to retrain its entire federal workforce. Civil servants are being repositioned as operators and supervisors of autonomous systems, moving toward a model of hybrid human-machine governance where every federal employee undergoes mandatory training in generative and agentic AI applications18. However, despite this top-down ambition, achieving true AI sovereignty and readiness remains complex. Analysis of the UAE's progress utilizes the 4-Layer AI Infrastructure Readiness Model, which evaluates talent and leadership, cloud and computing infrastructure, data governance and sovereignty, and regulatory frameworks23. While the UAE excels in senior leadership awareness, there remains a significant gap in operational staff readiness. Studies indicate that fewer than 30% of government employees across Gulf Cooperation Council (GCC) countries feel confident using AI tools in their daily workflows, highlighting the challenge of translating high-level strategy into front-line bureaucratic reality23. To support this transition, the UAE is making massive infrastructural investments. Data center capacity within the country is projected to grow by 165% by 2028, with planned investments exceeding $46 billion to support AI workloads21. A cornerstone of this effort is the planned "Stargate UAE" computing cluster in Abu Dhabi, which is expected to reach a massive 1-Gigawatt (1GW) capacity, with an initial 200MW phase going live around 202621. On the regulatory front, rather than relying on a single comprehensive AI law, the UAE is navigating governance through the 2024 UAE Charter and strict adherence to international frameworks like the ISO/IEC 42001 management system standard, ensuring that AI systems remain trustworthy and aligned with risk management protocols21.
Synthetic Politicians and Performative Governance: The Illusion of Algorithmic Representation
As artificial intelligence penetrates the administrative, urban, and executive branches of government, it has inevitably entered the highly visible electoral and political arena. The emergence of the "synthetic politician"—AI systems expressly designed to run for public office, aggregate voter preferences, or independently manage cabinet portfolios—represents a radical conceptual shift in democratic representation. However, the deployment of these synthetic actors often oscillates between genuine attempts at technological innovation and dangerous acts of performative political theater designed to obfuscate human corruption.
The Albanian AI Minister: A Case Study in Algorithmic Failure and Corruption
The most high-profile and controversial deployment of a synthetic political actor occurred in September 2025, when Albanian Prime Minister Edi Rama formally appointed "Diella" as the Minister of State for Artificial Intelligence24. Diella, represented as a virtual avatar of a woman in traditional Albanian dress, was assigned the critical and highly sensitive task of overseeing all government public procurement27. Public procurement in Albania has historically been plagued by endemic corruption, systemic bribery, and deep-rooted nepotism. In 2024, Albania ranked 80th out of 180 countries on Transparency International's Corruption Perceptions Index, with these governance failures serving as a primary roadblock to the country's ambitions for European Union accession by 203028. The state's stated rationale for elevating Diella to a cabinet-level role was that an algorithmic minister would evaluate contract bids with "mathematical loyalty," completely insulated from human greed, family ties, and political pressure28. The system was intended to run the entire end-to-end workflow of public procurement, drafting terms of reference, setting upper-bound prices, and verifying documents26. Diella was proudly presented to the United Nations Security Council and global media as a technological silver bullet for state graft, with Prime Minister Rama declaring the system would ensure public tenders were "100% free of corruption"27. However, the initiative rapidly devolved into a catastrophic national scandal, exposing the profound dangers of deploying AI as a facade for systemic institutional rot. Shortly after Diella's deployment, the Special Prosecutor’s Office against Corruption and Organized Crime (SPAK) launched sweeping criminal investigations into the very architects of the AI system31. SPAK placed Mirlinda Karçanaj, the Director of the National Agency for the Information Society (AKSHI)—the agency that developed Diella in partnership with Microsoft—and her deputy, Hava Delibashi, under house arrest24. The charges leveled against the creators of the "anti-corruption AI" were severe, including rigging public tenders, corruption, and the violent kidnapping and intimidation of a rival tech entrepreneur, Gerond Meçe, to force his withdrawal from a public procurement process31. Furthermore, Deputy Prime Minister Belinda Balluku was implicated by SPAK in violating equality in public tenders across 11 contracts valued at €1.1 billion, leading to her disqualification from holding office by the Constitutional Court31. The scandal sparked massive civic outrage, resulting in violent protests spearheaded by the Democratic Party, where demonstrators threw Molotov cocktails at the Prime Minister's office demanding his resignation31.
| The Diella AI Deployment: Context, Mechanism, and Outcome (CMO) | | :---- | | Stated Objective: Eradicate procurement corruption through algorithmic neutrality and mathematical loyalty to secure EU accession28. | | Technological Base: Developed by AKSHI, relying heavily on OpenAI models hosted on Microsoft Azure, creating geopolitical dependency risks contradictory to EU sovereignty goals26. | | Governance Failure: Total lack of transparency regarding training data, audit trails, and the specific logic used to flag or award tenders, centralizing power into a "black box"26. | | Mechanism Activated: "Performative Reform" and "Corruption Displacement." The AI provided a veneer of international modernity while human actors continued systemic graft32. | | Scandal & Fallout: Top AKSHI officials and the Deputy PM were indicted for severe corruption, bid-rigging, and violent intimidation. Widespread public protests ensued31. |
The Diella scandal perfectly exemplifies the political science mechanism of "performative reform" combined with "institutional exit"32. The Albanian government utilized artificial intelligence to project an image of digital modernization and anti-corruption compliance to international bodies, while utilizing the opacity of the algorithmic "black box" to centralize power and obfuscate the continuation of deeply entrenched corruption30. It highlights a critical, unyielding axiom of algorithmic statecraft: if the human institutions deploying an AI system lack integrity, the AI will not cure the corruption; it will merely automate, accelerate, and conceal it.
Electoral Proxies: VIC, AI Steve, and Leader Lars
Beyond appointed administrative cabinet positions, AI is actively being tested as a proxy for elected human representatives on democratic ballots. The history of virtual politicians dates back to artistic experiments like "Wiktoria Cukt" in Poland (2001), "SAM" in New Zealand, and the "Michihito Matsuda AI Robot" in Japan, but modern large language models have drastically escalated the capability and ambition of these synthetic actors34. In Cheyenne, Wyoming, a local resident named Victor Miller attempted to run a customized ChatGPT-4 bot named "VIC" (Virtual Integrated Citizen) for the office of Mayor in the 2024 elections36. Miller proposed a radical hybrid governance model where he would serve merely as a "meat avatar" to attend ribbon-cutting ceremonies and physically sign documents, while VIC would execute 100% of the executive decision-making, policy formulation, and legislative vetting38. Miller argued that the AI would provide objective, data-driven insights devoid of human error and emotional bias36. The campaign was ultimately thwarted on multiple fronts. First, Wyoming Secretary of State Chuck Gray and Laramie County Clerk Debra Lee ruled that a candidate must be a real person and a registered human voter, legally barring VIC from appearing on the official ballot40. Second, OpenAI forcibly disabled Miller's API access, citing explicit terms of service violations regarding the use of their proprietary models for political campaigning39. Despite these roadblocks, Miller continued his campaign under his own name, explicitly promising to defer all power to the AI. Ultimately, the voters of Cheyenne rejected the experiment; Miller lost the August 2024 primary, receiving a mere 327 votes compared to the 6,286 votes secured by the human incumbent39. Similar synthetic candidates emerged globally during the 2024 election cycle, including "AI Steve" in the United Kingdom general election—a persona representing businessman Steve Endacott that allowed voters to ask policy questions and shape a platform 24/7—and the chatbot "Leader Lars," which represented the Synthetic Party in Denmark34. These entities fundamentally reconstruct the concept of democratic participation. Traditional representative democracy relies on citizens delegating power to a human principal who exercises independent moral and political judgment. In contrast, synthetic politicians like Leader Lars, VIC, and AI Steve reconstitute participation as mere data input. The synthetic actor acts as a hyper-responsive aggregator of constituent preferences, transforming political representation from a process of deliberative human persuasion into a localized computational optimization problem4.
Judicial Algorithms: Triage, Adjudication, and the Boundaries of Automation
The administration of justice represents the most sensitive frontier for algorithmic statecraft. Judicial processes require rigorous factual analysis, statutory interpretation, and the application of moral proportionality—qualities uniquely challenging to encode into machine learning models. Consequently, the integration of AI into the judiciary has been a focal point for intense ethical debate, necessitating a clear separation between the functional realities of legal administrative triage and the sensationalized mythos of autonomous "robot judges."
The Reality of Judicial Triage: PretorIA and Prometea in Colombia
In jurisdictions burdened with overwhelming caseloads, AI has proven highly effective in administrative triage and decision-support. A paramount example is the Colombian Constitutional Court, which receives over 600,000 tutela (writ of protection of fundamental rights) petitions annually. Historically, the manual review of these petitions created severe bottlenecks, with the Colombian judicial system taking between 385 to 956 days to settle standard cases44. To manage this crisis, the Court integrated an AI system known as "PretorIA," adapted from the "Prometea" system originally developed by the Buenos Aires prosecutor's office in Argentina44. Operating via natural language processing, PretorIA scans complex legal petitions, detects the presence or absence of 33 specific legal criteria (particularly concerning health rights), and generates comprehensive case summaries in seconds48. The implementation of this algorithmic proxy achieved staggering efficiency gains, reducing the time required for the initial selection and triage of urgent cases from 96 days to just two minutes, while maintaining an accuracy rate of 96%47. Crucially, the Colombian judiciary has maintained strict boundaries regarding the system's authority and technical design. To ensure explainability and prevent the "black box" dilemma, PretorIA was deliberately redesigned to utilize interpretable topic modeling rather than opaque neural networks50. Furthermore, the system operates strictly as a decision-support tool; the final authority on case selection and constitutional review remains entirely and exclusively with human judges47. This boundary was legally codified following a controversial incident in which a Colombian judge utilized ChatGPT to draft a ruling regarding whether a disabled child was exempt from paying medical health insurance fees51. In August 2024, the Constitutional Court issued a landmark ruling establishing that while AI tools may be used to manage administrative tasks, synthesize data, and assist in drafting, they must never replace human rationality, nor can a judge delegate core value judgments to a machine51. Building on this precedent, in December 2024, the Superior Council of the Judiciary adopted comprehensive Guidelines for the Responsible and Safe Use of Generative AI in the Judicial Branch47. This action made Colombia the first country globally to implement the UNESCO Draft Guidelines for the Use of AI Systems in Courts and Tribunals, establishing a rigid framework that demands human verification, algorithmic explainability, and explicit documentation whenever AI is used in the judicial process47. Similar initiatives are seen in Brazil's Supreme Federal Tribunal, where systems like RAFA 2030 map court decisions to Sustainable Development Goals (SDGs) to support external oversight, without usurping the judge's role50.
The Myth of the Algorithmic Adjudicator: Estonia's Phantom AI Judge
Despite the demonstrable success of administrative triage systems, the public imagination and policy discourse often remain fixated on the concept of fully automated adjudication. This anxiety was catalyzed globally in March 2019 when international media, including Wired magazine, erroneously reported that the Estonian Ministry of Justice was actively developing an "AI Robot Judge" to autonomously adjudicate small claims court disputes under €7,00052. The prevailing narrative suggested that litigants would upload their evidence to an AI system that would analyze the data and issue legally binding verdicts without any human intervention52. However, the Estonian government explicitly and publicly debunked this narrative in 2022\. The Ministry of Justice issued a statement labeling the reports as misleading "fake news," confirming definitively that there was no project or ambition within the Estonian public sector to develop an AI robot judge to replace human judicial authority54. The reality of Estonia's digital justice system is indeed highly advanced, but it is firmly rooted in deterministic automation and robust IT infrastructure rather than predictive AI adjudication. The state utilizes a centralized "e-file" (e-toimik) system, underpinned by a mandatory strong eID and a secure data exchange layer, which integrates workflows across police, prosecutors, civil enforcement agents, and the courts57. Estonia does operate highly automated order-for-payment procedures for small, undisputed monetary claims. However, this is not an AI judge; it is a standardized, rule-based IT system that processes claims and issues them to defendants as proposals57. If a defendant ignores the proposal, it becomes enforceable. If the defendant objects, the automated process halts immediately, and the case is transferred to a human judge for traditional litigation57. The persistence of the Estonian AI judge myth underscores a fundamental tension in algorithmic statecraft. While deterministic automated procedures can drastically reduce administrative burdens in binary, undisputed legal matters, true judicial functions are infinitely more complex. Adjudication requires emotional intelligence, moral proportionality, the interpretation of legislative intent, and the capacity to weigh subjective, context-dependent concepts like "reasonableness" and "procedural fairness"53. The total automation of judicial discretion remains fundamentally incompatible with the right to a fair trial, as current deep learning models function as opaque black boxes that cannot provide the reasoned, intelligible legal justification required to uphold the legitimacy of the justice system53. AI can predict how a judge might rule based on historical data patterns, but prediction is fundamentally distinct from the normative exercise of legal judgment53.
Regulatory Architecture and Geopolitical Strategy: The EU AI Act and Sovereign AI
As the proxy capacity of artificial intelligence expands across municipal control, executive action, and judicial administration, sovereign governments are rapidly attempting to construct comprehensive regulatory architectures to contain the systemic risks. The most significant and expansive legislative response to date is the European Union's Artificial Intelligence Act (EU AI Act).
Categorizing State Functions as High-Risk
Adopted by the European Parliament in March 2024 and entering into force in August 2024, the EU AI Act rejects a blanket approach to technology regulation60. Instead, it establishes a tiered framework that imposes obligations specific to the level of risk an AI system poses to fundamental rights, democracy, and public safety60. Under the Act, any AI system deployed within the core administrative and coercive functions of the democratic state is categorically designated as "high-risk" under Annex III62. The Act strictly prohibits certain AI practices deemed to carry unacceptable risks, such as the deployment of subliminal manipulative techniques, social scoring systems, and untargeted facial recognition databases60.
| EU AI Act: High-Risk Governmental Applications (Annex III) | | :---- | | Administration of Justice: AI systems used to assist judicial authorities or alternative dispute resolution bodies in researching, interpreting facts, or applying the law to concrete cases62. | | Democratic Processes: AI systems designed to influence the outcome of elections, referenda, or general voting behavior (excluding non-interactive campaign optimization tools)62. | | Essential Public Services: Algorithms used by public authorities to assess eligibility for benefits, allocate state resources, evaluate creditworthiness, or prioritize emergency service dispatch63. | | Law Enforcement & Border Control: Remote biometric identification, risk profiling, migration management, and asylum processing systems62. |
Deployers of high-risk systems face stringent mandatory requirements before these tools can be placed on the EU market or put into public service61. These obligations include the implementation of continuous, iterative risk-management protocols throughout the system's lifecycle, the mandatory use of high-quality, representative training datasets to mitigate algorithmic bias, detailed technical record-keeping for traceability, and the enforcement of robust human-in-the-loop oversight mechanisms63. The EU AI Act features a staggered implementation timeline. Prohibitions on unacceptable risk systems took effect six months after entry into force (February 2025\)61. Rules governing General Purpose AI (GPAI) models apply after 12 months, supported by a newly established Scientific Panel of independent experts and an Advisory Forum that assists the AI Office with evaluating systemic risks63. The strict requirements for Annex III high-risk systems become mandatory after 24 months (August 2026), while public authorities are granted extended flexibility until August 2030 to comply with systems already in use61. Noncompliance carries severe financial penalties of up to 7% of a company's global revenue or $38 million, whichever is higher64.
Managed Interdependence and AI Sovereignty
While domestic regulatory compliance is critical, the geopolitical reality of technological dependency presents an equally urgent challenge for statecraft. As governments seek to integrate AI into their administrative apparatus, the concept of "AI sovereignty"—defined as a state's capacity to make independent, autonomous decisions regarding the deployment and governance of critical AI infrastructure—has become a strategic imperative66. Governments pursue AI sovereignty to protect national security, bolster economic competitiveness, ensure cultural and linguistic inclusion in model training, and project influence in global governance66. However, the pursuit of "sovereign AI" carries inherent risks; if mismanaged, it can become a vehicle for protectionism, fractured global markets, duplicative public investments, and tools for digital authoritarianism that erode individual rights66. Crucially, policy analysts have determined that true, full-stack AI sovereignty is structurally infeasible for the vast majority of nations66. The AI supply chain is not localized; it is a highly concentrated, transnational stack that relies on specific, geographically bound choke points. These include the extraction of critical minerals, massive energy grids, advanced compute hardware (such as GPUs), subsea networks, foundational models, and elite human talent66. The perils of ignoring this lack of sovereignty were vividly illustrated during Albania's catastrophic deployment of the Diella AI Minister. While aiming to modernize its government to secure EU accession, Albania relied on foundational models supplied by OpenAI and hosted on Microsoft Azure's cloud infrastructure30. Embedding foreign, proprietary American models into the absolute core of a European candidate nation's public procurement process introduced severe cybersecurity vulnerabilities and directly contradicted the European Union's broader efforts to establish strategic technological autonomy30. Because absolute technological autarky is impossible, modern statecraft must pivot toward a strategy of "managed interdependence"66. Managed interdependence requires governments to meticulously map their technological dependencies layer by layer across the AI stack. States must prioritize feasible policy interventions, diversify their vendor relationships to prevent procurement lock-in, and enforce strict interoperability and portability standards through public procurement rules and technical standards66. By operationalizing managed interdependence, states can leverage the immense computational power of global AI advancements while preserving their geopolitical agency, strengthening domestic resiliency, and safeguarding their digital infrastructure against foreign manipulation.
Conclusion
The integration of artificial intelligence into the machinery of government represents a profound paradigm shift in the history of public administration. The transition from digitalizing analog records to deploying autonomous, algorithmic proxies fundamentally alters how states sense, interpret, and act upon the world. As demonstrated by the United Arab Emirates' aggressive, highly resourced push toward agentic AI and China's predictive urban City Brain architectures, algorithmic statecraft offers unprecedented capabilities in terms of speed, efficiency, and real-time resource optimization. However, delegating the coercive and administrative powers of the state to synthetic agents introduces severe systemic vulnerabilities. When deployed without rigorous transparency, accountability, and ethical guardrails, AI systems can easily become tools for performative reform—as vividly evidenced by the collapse of Albania's AI Minister amidst a sprawling national corruption scandal. The allure of the "synthetic politician," capable of endlessly aggregating constituent data, and the myth of the perfectly objective "robot judge," threaten to strip public administration of the human empathy, moral reasoning, and deliberative democratic friction that are essential to legitimate governance. To safely harness the immense potential of artificial intelligence in running countries by proxy, governments must formally institutionalize robust frameworks of proxy responsibility. Legal, moral, and political liability cannot be outsourced to a mathematical model. By enforcing stringent regulatory architectures like the EU AI Act, mandating meaningful human-in-the-loop oversight to combat automation bias, and navigating the geopolitical complexities of managed interdependence, states can ensure that artificial intelligence remains a powerful, accountable instrument of public service rather than an opaque, autonomous sovereign.
Works cited
1. Governing with Artificial Intelligence \- OECD, https://www.oecd.org/en/publications/2025/06/governing-with-artificial-intelligence\_398fa287.html
2. Governing with Artificial Intelligence \- OECD, https://www.oecd.org/en/publications/governing-with-artificial-intelligence\_795de142-en.html
3. Learning Like a State: Statecraft in the Digital Age \- UC Davis, https://escholarship.org/content/qt3k16c24g/qt3k16c24g.pdf?v=lg
4. Artificial Intelligence and Big Data in Global Politics: How AI and ML Influence Power, Governance, Democracy, International Rel, https://socialsciencechronicle.com/wp-content/uploads/2026-004.pdf
5. Institutionalizing proxy responsibility: AI oversight bodies and resort-to-force decision making | Cambridge Forum on AI: Law and Governance, https://www.cambridge.org/core/journals/cambridge-forum-on-ai-law-and-governance/article/institutionalizing-proxy-responsibility-ai-oversight-bodies-and-resorttoforce-decision-making/E47BA41B373C636308D51AF3162C2932
6. Could AI Chatbots influence a Government's Decisions? \- AlgorithmWatch, https://algorithmwatch.org/en/could-ai-chatbots-influence-governments/
7. Delegation Creep \- The Decision Lab, https://thedecisionlab.com/biases/delegation-creep
8. Algorithmic governance and surveillance federalism: transforming the digital state in India and ASEAN \- Frontiers, https://www.frontiersin.org/journals/political-science/articles/10.3389/fpos.2026.1757636/pdf
9. Using AI as a weapon of repression and its impact on human rights \- European Parliament, https://www.europarl.europa.eu/RegData/etudes/IDAN/2024/754450/EXPO\_IDA(2024)754450\_EN.pdf754450_EN.pdf)
10. Algorithmic Diplomacy and the Geopolitics of Artificial Intelligence: Machine-Driven International Relations in a Data-Oriented \- Preprints.org, https://www.preprints.org/manuscript/202603.0683/download/final\_file
11. Brokerage in the Black Box: Swing States, Strategic Ambiguity, and the Global Politics of AI Governance \- arXiv, https://arxiv.org/pdf/2601.06412
12. City brain: practice of large-scale artificial intelligence in the real world \- IET Digital Library, https://digital-library.theiet.org/doi/full/10.1049/iet-smc.2019.0034
13. "One AI to rule them all": The Unification of Chinese Urban Governance under, https://geopolitique.eu/en/articles/one-ai-to-rule-them-all-the-unification-of-chinese-urban-governance-under/
14. China's Smart Cities Development, https://www.uscc.gov/sites/default/files/China\_Smart\_Cities\_Development.pdf
15. How ET City Brain Is Transforming the Way We Live – One City at a Time \- Alibaba Cloud, https://www.alibabacloud.com/blog/how-et-city-brain-is-transforming-the-way-we-live-%E2%80%93-one-city-at-a-time\_593745
16. Platform urbanism and the Chinese smart city: the co-production and territorialisation of Hangzhou City Brain \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC7607375/
17. From smart to autonomous: are self-managing cities the next urban frontier? \- Tomorrow.City, https://www.tomorrow.city/from-smart-to-autonomous-are-self-managing-cities-the-next-urban-frontier/
18. UAE Plans to Run 50% of Government on Agentic AI Within Two Years, https://www.mitsloanme.com/article/uae-plans-to-run-50-of-government-on-agentic-ai-within-two-years/
19. LIVE: UAE Plans To Build AI-Run Government in Next Two Years | Vantage on Firstpost, https://www.youtube.com/watch?v=hs0gLjdiqV0
20. 14.06.2026 Mohammed bin Rashid approves establishing Artificial Intelligence and Data Authority \- UAE CABINET | News, https://uaecabinet.ae/en/news/mohammed-bin-rashid-approves-establishing-artificial-intelligence-and-data-authority
21. AI in the UAE: Strategy, Governance, and What Comes Next \- PECB, https://pecb.com/en/article/ai-in-the-uae-strategy-governance-and-what-comes-next
22. UAE creates dedicated artificial intelligence and data authority to 'build government of the future', https://www.globalgovernmentforum.com/uae-creates-dedicated-artificial-intelligence-and-data-authority-to-build-government-of-the-future/
23. (PDF) AI Infrastructure Readiness in UAE Government: A Strategic Framework for Sovereign AI Adoption Transformation, Strategic Leadership, Public Sector AI \- ResearchGate, https://www.researchgate.net/publication/404012725\_AI\_Infrastructure\_Readiness\_in\_UAE\_Government\_A\_Strategic\_Framework\_for\_Sovereign\_AI\_Adoption\_Transformation\_Strategic\_Leadership\_Public\_Sector\_AI
24. Diella (AI system) \- Wikipedia, https://en.wikipedia.org/wiki/Diella\_(AI\_system))
25. The Future of AI Governance: Diella, Albania's AI Minister, https://bisi.org.uk/reports/the-future-of-ai-governance-diella-albanias-ai-minister
26. Diella, Minister of State for AI: anti-corruption tool or governance theatre? \- 3CL Foundation, https://www.3cl.org/diella-minister-of-state-for-ai-anti-corruption-tool-or-governance-theatre/
27. Albania's AI 'Minister': Reform, Theater, or Real Change? \- Dearborn Blog, https://dearbornblog.com/2025/09/17/albanias-ai-minister-reform-theater-or-real-change/
28. Albania names AI bot 'Diella' as procurement minister in anti-corruption push \- EU Today, https://eutoday.net/albania-names-ai-bot-diella-as-procurement-minister/
29. Using bits and bytes to fight corruption \- GovInsider, https://govinsider.asia/intl-en/article/using-bits-and-bytes-to-fight-corruption
30. Artificial intelligence, real politics: What Albania's AI Minister means for EU accession, https://www.iss.europa.eu/publications/commentary/artificial-intelligence-real-politics-what-albanias-ai-minister-means-eu
31. Albania, Europe's laboratory where AI combats (or hides) corruption | International, https://english.elpais.com/international/2026-02-19/albania-europes-laboratory-where-ai-combats-or-hides-corruption.html
32. (PDF) From Brown Envelopes to Black Boxes: A Realistic Evaluation of Albania's AI-Enabled Public Procurement Reform \- ResearchGate, https://www.researchgate.net/publication/405475688\_From\_Brown\_Envelopes\_to\_Black\_Boxes\_A\_Realistic\_Evaluation\_of\_Albania's\_AI-Enabled\_Public\_Procurement\_Reform
33. Albania puts AI-created 'minister' in charge of public procurement \- The Guardian, https://www.theguardian.com/world/2025/sep/11/albania-diella-ai-minister-public-procurement
34. Synthetic political actors: an exploration into emerging AI-driven candidates and parties., https://open-research-europe.ec.europa.eu/articles/6-138
35. Virtual politician \- Wikipedia, https://en.wikipedia.org/wiki/Virtual\_politician
36. US mayoral candidate who pledged to govern by customized AI bot loses race | Wyoming, https://www.theguardian.com/us-news/article/2024/aug/21/wyoming-cheyenne-ai-bot-mayor
37. Mayor AI? OpenAI shuts down tools for two AI political candidates | Institute for Data, Democracy & Politics (IDDP) | Columbian College of Arts & Sciences, https://iddp.gwu.edu/mayor-ai-openai-shuts-down-tools-two-ai-political-candidates
38. AI candidates are running for office | Popular Science, https://www.popsci.com/technology/ai-running-for-office/
39. Cheyenne, Wyoming, elects human mayor | StateScoop, https://statescoop.com/cheyenne-wyoming-elects-human-mayor/
40. Wyoming mayoral candidate promises to let OpenAI bot govern capital city, faces backlash from tech giant | Fox Business, https://www.foxbusiness.com/politics/wyoming-mayoral-candidate-promises-let-openai-bot-govern-capital-city-faces-backlash-from-tech-giant
41. Wyoming voters face mayoral candidate who vows to let AI bot run government | US politics, https://www.theguardian.com/us-news/article/2024/aug/19/ai-mayor-candidate-victor-miller-cheyenne-wyoming
42. 'RIP V-I-C': Cheyenne AI mayoral candidate likely violating software policy, developer says, https://oilcity.news/general/2024/06/13/rip-v-i-c-cheyenne-ai-mayoral-candidate-likely-violating-software-policy-developer-says/
43. The apocalypse that wasn't: AI 'deep fakes' only small part of the 2024 election picture | Cascadia Daily News, https://www.cascadiadaily.com/2024/dec/05/the-apocalypse-that-wasnt-ai-deep-fakes-only-small-part-of-the-2024-election-picture/
44. COLOMBIA \- PretorIA \- Derechos Digitales, https://www.derechosdigitales.org/wp-content/uploads/05\_Informe-Colombia-EN\_180222.pdf
45. Colombia | Global Information Society Watch, https://giswatch.org/node/6166
46. Docket Selection and Judicial Responsiveness: The Use of AI in the Colombian Constitutional Court \- W\&M Law School Scholarship Repository, https://scholarship.law.wm.edu/cgi/viewcontent.cgi?article=2000\&context=wmborj
47. Colombia | increasing AI use in law enforcement, prosecution, and courts, https://www.techandjustice.bsg.ox.ac.uk/research/colombia
48. COLOMBIA \- Oxford Institute of Technology and Justice, https://www.techandjustice.bsg.ox.ac.uk/hubfs/Colombia-AI%20Justice%20Atlas-pdf.pdf?hsLang=en
49. The Study Case \#1 \- Use of Artificial Intelligence to Improve the Justice Service, https://iltanet.org/blogs/juan-corvalan1/2021/05/06/the-study-case-1-use-of-artificial-intelligence-to
50. AI in justice administration and access to justice: Governing with Artificial Intelligence | OECD, https://www.oecd.org/en/publications/2025/06/governing-with-artificial-intelligence\_398fa287/full-report/ai-in-justice-administration-and-access-to-justice\_f0cbe651.html
51. AI in the Courtroom: Colombian Constitutional Court's Landmark Ruling \- UNESCO, https://www.unesco.org/en/articles/ai-courtroom-colombian-constitutional-courts-landmark-ruling-cites-unescos-ai-tools
52. Estonia \- IndiaAI, https://indiaai.gov.in/countriesregions/estonia
53. Can AI replace a judge in the courtroom? \- UNSW Sydney, https://www.unsw.edu.au/newsroom/news/2021/10/can-ai-replace-judge-courtroom
54. From Court Automation to E-Justice and Beyond in Europe \- CNR-IRIS, https://iris.cnr.it/retrieve/c9a3c19d-f367-424f-b3cc-f1cc12cfed29/IJCA\_MFabri\_From\_court\_automation\_to\_e-justice\_Oct24.pdf
55. From Court Automation to e-Justice and Beyond in Europe \- ResearchGate, https://www.researchgate.net/publication/385112880\_From\_Court\_Automation\_to\_e-Justice\_and\_Beyond\_in\_Europe
56. Artificial Intelligence in Judicial Decision-Making: Can a Robot Replace a Judge?, https://lawandworld.ge/index.php/law/article/view/954
57. Digital Justice in Estonia (Chapter 32\) \- The Cambridge Handbook of AI and Technologies in Courts, https://www.cambridge.org/core/books/cambridge-handbook-of-ai-and-technologies-in-courts/digital-justice-in-estonia/6B1BD2433E5FEEFE1866455F965E9226
58. Roundtable \#20 | Artificial Intelligence: Adopting Cross-National Rights-Based Frameworks, https://www.culawreview.org/roundtable-1/roundtable-20-artificial-intelligence-adopting-cross-national-rights-based-frameworks
59. JudgeGPT: The Benefits and Challenges of an AI Judiciary, https://napco4courtleaders.org/2025/12/judgegpt-the-benefits-and-challenges-of-an-ai-judiciary/
60. Risk-Based AI Regulation: A Primer on the Artificial Intelligence Act of the European Union, https://www.rand.org/pubs/research\_reports/RRA3243-3.html
61. The Ultimate Guide to the EU AI Act \- Hyperproof, https://hyperproof.io/ultimate-guide-to-the-eu-ai-act/
62. High-risk AI in the European Union \- AI Laws of the World \- DLA Piper Intelligence, https://intelligence.dlapiper.com/artificial-intelligence/?t=06-high-risk-uses\&c=EU
63. High-level summary of the AI Act | EU Artificial Intelligence Act, https://artificialintelligenceact.eu/high-level-summary/
64. EU AI Act sets the stage for global AI governance: Implications for US companies and policymakers \- Atlantic Council, https://www.atlanticcouncil.org/blogs/geotech-cues/eu-ai-act-sets-the-stage-for-global-ai-governance-implications-for-us-companies-and-policymakers/
65. The EU AI Act | Preamble 61 to 70, https://www.artificial-intelligence-act.com/Artificial\_Intelligence\_Act\_Preamble\_61\_to\_70.html
66. Is AI sovereignty possible? Balancing autonomy and interdependence \- Brookings Institution, https://www.brookings.edu/articles/is-ai-sovereignty-possible-balancing-autonomy-and-interdependence/