PROJECT: ConstAI Database
Justification of the Categorisation Framework of the Codebook of the ConstAI Database
Developed and maintained by the Algorithmic Constitutionalism LENDULET/Momentum Research Group
The present document provides an academic justification for the categorisation framework adopted in the Codebook of the ConstAI Database, developed within the Algorithmic Constitutionalism Lendület/Momentum Research Project. The codebook structures the global, open-access database that forms the central empirical output of the project, cataloguing normative acts, soft law instruments, and judicial case law that address the intersection of artificial intelligence and three selected fundamental rights: the right to a fair trial, freedom of expression, and the right to a healthy environment.
The selection of these three thematic pillars is theoretically grounded. Algorithmic constitutionalism describes regulatory and interpretive efforts to accommodate non-human actors possessing unprecedented human-like characteristics within legal systems designed for human persons (Pollicino and de Gregorio 2021; Pollicino and Paolucci 2025). The codebook translates this conceptual framework into an operational taxonomy that can structure comparative, cross-jurisdictional legal research. The justification below proceeds in three parts, addressing the rationale for the overarching thematic structure, the internal subdivision of each major topic, and the cross-cutting coding principles.
The choice of the right to a fair trial as the first major topic reflects the centrality of procedural justice in constitutional theory and the rapidly accelerating deployment of AI within judicial and administrative proceedings worldwide. Scholarly literature has extensively documented the adoption of algorithmic tools across multiple stages of judicial processes—from pre-trial risk assessment and recidivism scoring to automated admissibility screening (Volokh 2019; Engel et al. 2025), anonymisation of court documents, and AI-generated written records (Selçuk et al. 2025; Zalnieriute and Limante 2026). The Estonian model, the COMPAS system in the United States, and the PretorIA software in Colombia represent amongst many others paradigmatic examples of the operational reality that the codebook must capture.
The codebook’s substantive subtopics under Major Topic 1 (codes 101–113) map onto the procedural dimensions of the right to a fair trial. Subtopics 101–110 track the core components of the guarantee, while 111–113 respond specifically to the rapid proliferation of generative AI in litigation. The subtopics are as follows:
100: General – Right to a Fair Trial. Applies where a document genuinely spans two or more subtopics of this major topic with no single predominant concern (for example an overarching principle declaration addressing several fair-trial dimensions at once). See Section 3 for the exclusivity rule.
101: Profiling, Prediction and Fair Proceedings. AI used to profile or predict, and its effect on the fairness of proceedings.
102: Evaluation, Classification and Non-Discrimination in Proceedings. Algorithmic evaluation and classification of persons, and the prohibition of discrimination in a procedural context.
103: Biometric Identification in Judicial Contexts. Biometric identification deployed within judicial or investigative proceedings.
104: Facial Recognition, Emotion Inference and Biometric Categorisation. Facial recognition, emotion inference and biometric categorisation bearing on fair-trial guarantees.
105: High-Risk AI Systems – Transparency and Human Oversight in Justice. Transparency and human-oversight requirements for high-risk AI used in the administration of justice.
106: High-Risk AI Systems – Fundamental Rights Impact Assessment. Fundamental-rights impact assessment duties for high-risk justice-sector AI.
107: Remedy, Complaint and Explanation in AI-Assisted Decisions. Rights to a remedy, to complain and to an explanation where decisions are assisted by AI.
108: High-Risk AI Systems – Bias Detection and Personal Data in Justice. Bias detection and personal-data safeguards for high-risk AI in the justice sector.
109: AI Regulatory Sandbox and Testing in Justice Contexts. Controlled testing environments for AI systems used in justice contexts.
110: High-Risk AI Systems – Protection of Minors and Vulnerable Groups. Protection of minors and other vulnerable groups in relation to high-risk justice-sector AI.
111: Misuse of Generative AI in Legal Proceedings. Case law in which parties, counsel, experts or judges have relied on fabricated or hallucinated AI-generated citations and submissions—a phenomenon documented across nearly 800 incidents in at least twenty-five jurisdictions by late 2025 (Charlotin 2025; Magesh et al. 2025).
112: Normative Restrictions on the Use of Publicly Available AI Tools. Binding instruments that prohibit or restrict the use of publicly available AI tools in judicial proceedings.
113: Normative Rules on the Use of Publicly Available AI Tools in Court Proceedings. The broader body of normative rules that govern such use.
199: Other – AI and the Right to a Fair Trial. Applies where a document falls within this major topic but cannot be assigned to any specific subtopic (typically appropriations or budgetary measures, honorary or symbolic resolutions, and instruments that do not substantively engage any fair-trial dimension). See Section 3.
This taxonomy mirrors the risk-based architecture of the EU Artificial Intelligence Act, which classifies AI systems used in the administration of justice and democratic processes as high-risk, thereby triggering the most stringent regulatory requirements (Palmiotto 2025; Mizaras et al. 2026).
The inclusion of a dedicated residual subcategory and a general combination code follows established practice in comparative legal research databases, where full coverage without over-coding requires both a catch-all category and a multi-topic aggregator (Sekwenz et al. 2025). The residual code is applied uniformly across all three major topics as 199 (fair trial), 299 (freedom of expression) and 399 (healthy environment), and the general combination code correspondingly as 100, 200 and 300. To ensure cross-coder consistency, the two code families are governed by an explicit and mutually exclusive rule. The General code (x00) is reserved for documents that genuinely span two or more subtopics of the same major topic with no single predominant concern—typically overarching national principle declarations, ethics charters or framework instruments that address several fundamental-rights dimensions at once. The residual code (x99) applies only to documents that fall within the major topic yet cannot be assigned to any specific subtopic—characteristically appropriations and budgetary measures, honorary or symbolic resolutions, and instruments that do not substantively engage any of the codified fundamental-rights dimensions. Where exactly one subtopic predominates, that specific subtopic code must be assigned; neither x00 nor x99 may be used as a default for coder uncertainty. A recurrent cluster of documents establishing AI advisory councils, commissions and task forces—institutional governance bodies without a substantive rights-engaging mandate—is captured by a dedicated subtopic (213, AI Advisory Bodies, Councils and Governance Institutions) rather than left in the residual category, so that the residual code retains its intended narrowness.
Freedom of expression occupies the second major topic because AI-related threats to the open public sphere represent one of the most extensively documented and judicially contested dimensions of the societal impact of AI-based technology'. The project proposal situates this concern within a wider phenomenon: AI-enabled computational propaganda, deep fake technology, algorithmic amplification of disinformation, and the use of bots to distort democratic discourse have each generated substantial legislative responses (Szentgáli-Tóth et al. 2023; Bassini 2025; De Gregorio and Pollicino 2025). The European Court of Human Rights acknowledged the transformative role of internet platforms for expressive activity in Delfi AS v. Estonia (2015), and the subsequent wave of platform regulation—from the EU Digital Services Act to national anti-disinformation legislation—reflects a sustained effort to balance the informational openness protected by freedom of expression with the harms facilitated by AI-driven content manipulation (Łabuz 2025).
The codebook’s twelve substantive subtopics under Major Topic 2 (codes 201–212) are structured along functional lines reflecting the primary dimensions identified in the constitutional scholarship. The subtopics are as follows:
200: General – Freedom of Expression. Applies where a document genuinely spans two or more subtopics of this major topic with no single predominant concern. See Section 3 for the exclusivity rule.
201: Manipulation, Disinformation and Distorted Behaviour. The political and electoral dimension of AI-driven influence operations.
202: Transparency, Accountability and AI–Human Interaction. Procedural safeguards—such as disclosure and labelling obligations—that mediate the user’s epistemic relationship with AI systems.
203: Generative AI, Deep Fakes and Freedom of Expression. Generative AI and deep fakes as a distinct constitutional concern affecting both expression and data protection.
204: Political Participation, Democracy and AI. Instruments protecting the integrity of the democratic public sphere.
205: Labour Relations, Collective Agreements and Expression Rights. The intersection of AI, labour relations and expression rights in the workplace.
206: Education, AI Literacy and Digital Skills for Expression. Education, AI literacy and the digital skills that underpin the exercise of expression.
207: Accessibility and Inclusive Access to Information. Accessibility and inclusive access to information and to AI systems.
208: Cybersecurity, Surveillance and Freedom of Expression. The impact of cybersecurity measures and surveillance on freedom of expression.
209: Personal Data, Privacy and Expression. The interaction of personal-data protection and privacy with expression rights.
210: National AI Strategies, Governance Frameworks and Freedom of Expression. National AI strategies and governance frameworks as they bear on transparency and expression.
211: High-Risk AI Systems – Transparency, Accountability and Expression. Transparency and accountability requirements for high-risk AI systems affecting expression.
212: AI Regulatory Sandbox, Testing and Expression. Controlled testing environments for AI systems bearing on freedom of expression.
213: AI Advisory Bodies, Councils and Governance Institutions. Instruments whose primary purpose is the creation, composition or mandate of a body tasked with advising on, overseeing or governing artificial intelligence — such as advisory councils, expert commissions, task forces, dedicated AI offices and inter-institutional governance structures. This subtopic captures the institutional dimension of AI governance: where the operative content of the instrument is the establishment of an institution rather than a substantive rule, the instrument is coded here rather than under a thematic subtopic or the General or residual categories, even where the body’s remit touches several fundamental rights.
299: Other – AI and Freedom of Expression. Applies where a document falls within this major topic but cannot be assigned to any specific subtopic. See Section 3.
Taken together, these subtopics ensure that the full range of AI–expression intersections documented in the literature is operationally covered.
The explicit cross-referencing instructions within the codebook—notably between 201 and 203 (deep fakes as manipulation versus expression), and between 209 and 202 (privacy and transparency)—reflect the doctrinal observation that fundamental rights frequently interact in the AI context, requiring coding decisions that are both principled and reproducible across coders.
The right to a healthy environment was selected as the third major topic for two complementary reasons. First, it belongs to the third generation of fundamental rights and has received comparatively less attention in the AI-constitutionalism literature, making it a high-value addition to the research database. Second, AI is simultaneously a significant contributor to environmental stress—through energy consumption, hardware production, and data centre operations—and a potential instrument of environmental protection through monitoring, predictive modelling, and resource optimisation (Vinuesa et al. 2020; Noorman et al. 2023; Li et al. 2025; Xiao et al. 2025). The constitutional and legal instruments that regulate this double role of AI vis-à-vis environmental rights constitute a rapidly growing corpus that has not yet been systematically indexed.
The eleven substantive subtopics under Major Topic 3 (codes 301–311) are organised strictly around the regulatory function that AI performs in relation to the right to a healthy environment, rather than around the economic sector in which it is deployed. This preserves the functional coding logic used in Major Topics 1 and 2. Sectoral context (for example energy, transport, agriculture or public health) is not a primary code but is recorded, where relevant, through the cross-referencing mechanism described in Section 3. The subtopics are as follows:
300: General – Right to a Healthy Environment. Applies where a document genuinely spans two or more subtopics of this major topic with no single predominant concern. See Section 3 for the exclusivity rule.
301: Environmental Monitoring, Assessment and Data. AI as a tool of environmental knowledge production: measurement, remote sensing, forecasting and environmental impact assessment.
302: National AI Strategies with an Environmental or Sustainability Dimension. The policy-level intersection, where national AI strategies address environmental or sustainability objectives.
303: Accessibility, Public Health and Environmental Well-being. AI directed at inclusive access and public-health-related environmental well-being, linking the right to a healthy environment with non-discrimination.
304: Energy Efficiency, Climate Change and AI’s Environmental Footprint. The quantitatively most significant dimension: the energy and resource footprint of AI systems and the regulatory responses to it, including the energy transition.
305: AI and Food and Resource Safety. AI functions bearing on food safety and the safety of environmental resources.
306: High-Risk Environmental AI Systems. AI systems classified as high-risk by reason of their environmental or environmental-health implications.
307: AI, Transport and Environmental Impact. The environmental function of AI in the transport sector (emissions, mobility optimisation, environmental monitoring of transport infrastructure), not transport regulation as such.
308: AI, Social Equity and Environmental Justice. The distributional and environmental-justice dimension, including the protection of vulnerable and historically excluded groups.
309: AI in Environmental Governance and Public Administration. Administrative, permitting and public-governance decisions relating to the environment that are supported or determined by AI.
310: AI Regulatory Sandboxes in the Environmental Domain. Controlled testing environments for environmental AI systems.
311: AI Accountability, Transparency and Environmental Oversight. Accountability, transparency and oversight mechanisms specific to environmental AI.
399: Other – AI and the Right to a Healthy Environment. Applies where a document falls within this major topic but cannot be assigned to any specific subtopic. See Section 3.
Codes 305, 306 and 310 are defined for completeness of the functional taxonomy and may not yet correspond to coded entries; their presence ensures that the framework can accommodate such instruments consistently as the corpus grows (Noorman et al. 2023; Li et al. 2025; Xiao et al. 2025).
Three cross-cutting principles govern the codebook's design and require specific justification.
First, the tripartite document typology—binding normative acts, soft law instruments, and judicial case law—reflects the project's foundational theoretical commitment to examining all three levels of legal adaptation identified in the proposal: normative rules, soft law, and interpretive case law. This taxonomy is standard in comparative constitutional law research (de Gregorio 2022; Pollicino and Paolucci 2025) and ensures that the database captures the full regulatory ecosystem rather than privileging hard law over the growing corpus of AI guidelines, codes of practice, and recommendations issued by international bodies such as the Council of Europe, the United Nations, and the OECD.
Second, the principle of primary-focus coding—under which each document receives a single major topic and a single subtopic reflecting its predominant concern—is methodologically necessary to ensure coding reliability and cross-coder consistency. Where documents genuinely span multiple subtopics of the same major topic, the 'General' (x00) codes provide a principled alternative to forced specificity, subject to the exclusivity rule set out above: the General code is not a repository for cases the coder finds difficult to assign, but is confined to instruments in which no single subtopic predominates. This approach follows established practice in large-scale legal content analysis (Sekwenz et al. 2025).
Third, the use of explicit cross-referencing instructions within the coding guide operationalises the recognition that AI-related fundamental rights concerns are deeply interconnected. Biometric identification, for instance, implicates both fair trial guarantees (Category 1) and freedom of expression (Category 2) when deployed for surveillance of political dissent. Similarly, energy poverty intersects with both environmental rights (Category 3) and the right to a fair trial when algorithmic utility management generates administrative decisions subject to judicial review. The cross-references embedded in the codebook allow coders to record primary classification while flagging secondary connections, thereby enabling subsequent relational analysis across the database. This mechanism also resolves a structural asymmetry between the three major topics. Whereas the subtopics of Major Topics 1 and 2 are organised consistently along functional lines—each subtopic corresponding to a distinct role that AI plays in relation to the right—several subtopics under Major Topic 3 are framed by economic sector (for example, food safety and transportation) rather than by function. To preserve comparability across the three pillars without restructuring the taxonomy, sectoral information under Major Topic 3 is treated as a secondary, cross-referenced attribute rather than as the primary classificatory axis: the predominant regulatory function of the AI system (environmental monitoring, energy and climate optimisation, environmental governance, and so forth) governs the primary subtopic code, while the sector in which it operates is recorded through the cross-referencing mechanism. Coding decisions under Major Topic 3 therefore follow the same function-based logic as the other two pillars, and sectoral granularity is retained for relational querying without fragmenting the primary code.
Bassini, M, ‘Speech without a Speaker: Constitutional Coverage for Generative AI Output?’ (2025) 21 European Constitutional Law Review 1.
Charlotin, D, AI Hallucination Cases Database (HEC Paris 2025) <https://www.damiencharlotin.com/hallucinations/>.
de Gregorio, G, Digital Constitutionalism in Europe (Cambridge University Press 2022).
De Gregorio, G and Pollicino, O, ‘The European Constitutional Way to Address Disinformation in the Age of Artificial Intelligence’ (2025) 26 German Law Journal 449.
Engel, C, Linhardt, L and Schubert, M, ‘Code Is Law: How COMPAS Affects the Way the Judiciary Handles the Risk of Recidivism’ (2025) 33 Artificial Intelligence and Law 383.
Łabuz, M, ‘A Teleological Interpretation of the Definition of DeepFakes in the EU Artificial Intelligence Act—A Purpose-Based Approach to Potential Problems With the Word “Existing”’ (2025) 17(1) Policy & Internet e435.
Li, P, Yang, J, Islam, M A and Ren, S, ‘Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models’ (2025) 68(7) Communications of the ACM 54.
Magesh, V, Surani, F, Dahl, M, Suzgun, M, Manning, C D and Ho, D E, ‘Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools’ (2025) 22 Journal of Empirical Legal Studies 216.
Mizaras, V, Jurcys, P, Fenwick, M, Earp, B D, Kozuka, S, Porsdam Mann, S and Dwivedi, Y K, ‘AI Judges in Courts of the Future and the Right to a Fair Trial in the ECHR’ in M Zalnieriute and A Limante (eds), The Cambridge Handbook of AI and Technologies in Courts (Cambridge University Press 2026) 419.
Noorman, M, Apráez Espinosa, B and Lavrijssen, S, ‘AI and Energy Justice’ (2023) 16(13) Energies 4922.
Palmiotto, F, ‘The AI Act Roller Coaster: The Evolution of Fundamental Rights Protection in the Legislative Process and the Future of the Regulation’ (2025) 16 European Journal of Risk Regulation 770.
Pollicino, O and De Gregorio, G, ‘Constitutional Law in the Algorithmic Society’ in H-W Micklitz and others (eds), Constitutional Challenges in the Algorithmic Society (Cambridge University Press 2021) 3.
Pollicino, O and Paolucci, F, ‘Digital Constitutionalism’ in L Floridi and M Taddeo (eds), A Companion to Digital Ethics (Wiley 2025) 313.
Sekwenz, M, Gsenger, R, Dahlgren, S and Wagner, B, ‘Doing Audits Right? The Role of Sampling and Legal Content Analysis in Systemic Risk Assessments and Independent Audits in the Digital Services Act’ (arXiv, 6 May 2025) arXiv:2505.03601.
Selçuk, S, Kurt Konca, N and Kaya, S, ‘AI-Driven Civil Litigation: Navigating the Right to a Fair Trial’ (2025) 57 Computer Law & Security Review 106136.
Szentgáli-Tóth, B, Gál, K, Kálmán, K and Toplak, J, ‘Freedom of Expression and Misinformation Laws During the COVID-19 Pandemic and the European Court of Human Rights’ (2023) 21(1) Lex Localis 213.
Vinuesa, R, Azizpour, H, Leite, I, Balaam, M, Dignum, V, Domisch, S, Felländer, A, Langhans, S D, Tegmark, M and Fuso Nerini, F, ‘The Role of Artificial Intelligence in Achieving the Sustainable Development Goals’ (2020) 11 Nature Communications 233.
Volokh, E, ‘Chief Justice Robots’ (2019) 68 Duke Law Journal 1135.
Xiao, T, Fuso Nerini, F, Matthews, H D, Tavoni, M and You, F, ‘Environmental Impact and Net-Zero Pathways for Sustainable Artificial Intelligence Servers in the USA’ (2025) 8(12) Nature Sustainability 1541.