AI Governance Literature Map (T1: ai-governance)

1. Short scope note

This note maps the academic and policy literature on AI governance frameworks, taxonomies, and major schools of thought from roughly 2018 onward. It prioritizes peer-reviewed papers, systematic reviews, research-institute reports, and primary-source framework documents used in the literature. It does not aim to provide a current legal-status memo on specific regulations, a deep review of frontier-lab internal practices, or a detailed treatment of empirical enforcement failures.

2. Numbered source list with titles, year, and direct URLs

S1. Corinne Cath, "Governing artificial intelligence: ethical, legal and technical opportunities and challenges" (2018) https://pmc.ncbi.nlm.nih.gov/articles/PMC6191666/

S2. Anna Jobin, Marcello Ienca, Effy Vayena, "The global landscape of AI ethics guidelines" (2019) https://arxiv.org/abs/1906.11668

S3. Jessica Fjeld, Nele Achten, Hannah Hilligoss, Adam Nagy, Madhulika Srikumar, "Principled Artificial Intelligence: Mapping Consensus in Ethical and Rights-Based Approaches to Principles for AI" (2020) https://dash.harvard.edu/bitstream/handle/1/42160420/HLS%20White%20Paper%20Final_v3.pdf

S4. Bernd W. Wirtz, Jan C. Weyerer, Benjamin J. Sturm, "The Dark Sides of Artificial Intelligence: An Integrated AI Governance Framework for Public Administration" (2020) https://econpapers.repec.org/RePEc:taf:lpadxx:v:43:y:2020:i:9:p:818-829

S5. Sandeep Reddy, Sonia Allan, Simon Coghlan, Paul Cooper, "A governance model for the application of AI in health care" (2020) https://pmc.ncbi.nlm.nih.gov/articles/PMC7647243/

S6. UNESCO, "Recommendation on the Ethics of Artificial Intelligence" (2021) https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence

S7. Matti Mäntymäki, Matti Minkkinen, Teemu Birkstedt, Mika Viljanen, "Putting AI Ethics into Practice: The Hourglass Model of Organizational AI Governance" (2022/2023 version) https://arxiv.org/abs/2206.00335

S8. Qinghua Lu, Liming Zhu, Xiwei Xu, Jon Whittle, Didar Zowghi, Aurelie Jacquet, "Responsible AI Pattern Catalogue: A Collection of Best Practices for AI Governance and Engineering" (2022/2023) https://arxiv.org/abs/2209.04963

S9. OECD, "AI Principles" (2019; updated 2024 page) https://www.oecd.org/en/topics/ai-principles.html

S10. OECD, "Advancing accountability in AI: governing and managing risks throughout the lifecycle for trustworthy AI" (2023) https://www.oecd.org/content/dam/oecd/en/publications/reports/2023/02/advancing-accountability-in-ai_753bf8c8/2448f04b-en.pdf

S11. NIST, Elham Tabassi, "Artificial Intelligence Risk Management Framework (AI RMF 1.0)" (2023) https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10

S12. Matthijs M. Maas, "Concepts in Advanced AI Governance: A Literature Review of Key Terms and Definitions" (2023) https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4612473

S13. Matthijs M. Maas, "Advanced AI Governance: A Literature Review of Problems, Options, and Proposals" (2023) https://matthijsmaas.com/publication/2023.-maas-4-literature-review/

S14. Simon Knight, Antonette Shibani, Nicole Vincent, "Ethical AI governance: mapping a research ecosystem" (2024) https://opus.lib.uts.edu.au/handle/10453/175681

S15. Amna Batool, Didar Zowghi, Muneera Bano, "AI governance: a systematic literature review" (2025) https://link.springer.com/article/10.1007/s43681-024-00653-w

S16. Paul Oudin, Teodora Groza, "The Governance of AI Companies: Reconciling Purpose with Profits" (2024) https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4972751

S17. Anka Reuel et al., "Open Problems in Technical AI Governance" (2024/2025 version) https://arxiv.org/abs/2407.14981

3. Evidence table

Claim	Evidence	Source refs	Confidence
1. AI governance is defined broadly, not just as regulation.	Across the literature, AI governance spans ethical principles, organizational controls, technical assurance, public policy, standards, and institutional oversight over the whole AI lifecycle. UNESCO explicitly treats AI ethics/governance as applying across research, design, development, deployment, use, maintenance, monitoring, and termination. NIST frames governance as practical risk management for organizations designing, developing, deploying, or using AI.	S6, S11, S15	High
2. A recurring baseline definition is socio-technical and multi-level.	Cath argues AI governance requires ethical, legal-regulatory, and technical responses together. Batool et al. synthesize the field into team, organization, industry, national, and international levels. Knight et al. treat governance as an ecosystem rather than a single control point.	S1, S14, S15	High
3. From 2018 onward, a first major family is principles-based / ethics-guideline governance.	Jobin et al. map a large corpus of ethics guidelines and show convergence around transparency, justice/fairness, non-maleficence, responsibility, and privacy. Fjeld et al. similarly map cross-document consensus in ethical and rights-based principles. This family is high-level and norm-setting rather than strongly operational.	S2, S3	High
4. A second major family is risk-based governance.	NIST AI RMF operationalizes governance through Govern/Map/Measure/Manage functions. OECD’s accountability report links trustworthy AI to risk management across the lifecycle. Batool et al. identify risk management as a dominant contemporary orientation.	S10, S11, S15	High
5. A third major family is lifecycle governance.	UNESCO explicitly states ethical questions pertain to all stages of the AI system lifecycle. OECD’s accountability report is explicitly lifecycle-based. The Hourglass model also ties governance requirements to the AI system lifecycle.	S6, S7, S10	High
6. A fourth major family is organizational / operational governance.	The Hourglass model focuses on environmental, organizational, and AI-system levels inside firms. Lu et al.’s pattern catalogue provides operational governance and engineering patterns. Reddy et al. give a concrete domain governance model for healthcare deployment.	S5, S7, S8	High
7. A fifth major family is sociotechnical / participatory governance.	Cath emphasizes governance beyond technical fixes and warns against narrow, industry-led agendas. UNESCO repeatedly stresses multi-stakeholder, multidisciplinary, inclusive governance. Knight et al. show governance in practice as a research ecosystem, not just a compliance checklist.	S1, S6, S14	High
8. A sixth major family, emerging more clearly after 2022, is advanced/frontier/model governance.	Maas distinguishes “advanced AI governance” as a field concerned with high-stakes future capabilities, governance options, and policy proposals. Reuel et al. define “technical AI governance” as technical analysis and tools supporting effective governance and provide a taxonomy of open problem areas.	S12, S13, S17	High
9. Corporate governance has become a distinct subfield intersecting with AI governance.	Oudin and Groza argue AI-company governance must reconcile public-purpose claims and profit incentives. The literature increasingly treats boards, governance structure, control rights, and corporate form as part of AI governance rather than downstream business details.	S16, S15	Medium-High
10. There is strong literature consensus that principle lists alone are insufficient.	Cath criticizes weak enforcement and conceptual ambiguity in ethics-only approaches. Jobin et al. show convergence on principles but not on implementation. Batool et al. conclude many studies lack robust mechanisms for who governs, when, and how.	S1, S2, S15	High
11. There is broad consensus that governance must be multi-stakeholder and cannot be reduced to engineers or regulators alone.	UNESCO expressly calls for multi-stakeholder, multidisciplinary, adaptive governance. OECD principles and accountability materials emphasize shared responsibility across actors. Batool et al. find governance roles distributed across organizational, national, and international levels.	S6, S9, S10, S15	High
12. There is broad consensus that governance should span the full AI lifecycle, not only deployment.	UNESCO, OECD, NIST, and the Hourglass model all anchor governance before deployment and after release, including monitoring and evaluation. This is one of the clearest cross-family points of agreement.	S6, S7, S10, S11	High
13. Human rights and public-interest framing are a major school of thought, especially in international and European-oriented literature.	UNESCO centers human dignity, rights, non-discrimination, and social justice. Fjeld et al. map rights-based approaches as a major strand. Cath argues governance agendas should include human rights and democratic values rather than only innovation and efficiency.	S1, S3, S6	High
14. A managerial / enterprise-governance school emphasizes organizational processes, accountability roles, and internal controls.	The Hourglass model, healthcare governance model, and pattern-catalogue work all focus on translating principles into organizational procedures, governance roles, decision gates, and documentation practices.	S5, S7, S8	High
15. A public-administration school emphasizes governing AI as a source of administrative risk and legitimacy problems.	Wirtz et al. develop an integrated governance framework for public administration centered on the “dark sides” of AI, including legitimacy, accountability, discrimination, privacy, and control concerns.	S4	Medium-High
16. A frontier/advanced-governance school disagrees with generic governance approaches by arguing that highly capable models create distinct governance problems.	Maas separates advanced AI governance from more general responsible-AI work by focusing on problem clarification, governance options, and policy proposals for high-stakes systems. Reuel et al. similarly argue some governance barriers are specifically technical and tied to advanced AI capabilities.	S12, S13, S17	High
17. Risk-based governance is influential, but the literature does not treat it as sufficient on its own.	OECD and NIST both use risk-management structures, yet UNESCO adds broader normative commitments, and Cath warns against reducing social and political issues to technocratic control. Batool et al. also note under-attention to ethics, human involvement, and broader governance adequacy.	S1, S6, S10, S11, S15	High
18. The literature increasingly links governance to accountability, audit, and impact assessment.	UNESCO recommends ethical impact assessment, oversight, auditability, and traceability. OECD’s accountability report is explicitly about governing and managing risks for trustworthy AI. Cath’s review highlights auditing and impact assessment as core governance mechanisms.	S1, S6, S10	High
19. The field remains terminologically unstable.	Maas’s definitional review is itself motivated by lack of clarity around “advanced AI governance” and related terms. Batool et al. also note literature fragmentation and uneven coverage of governance layers and artifacts.	S12, S15	High
20. The literature has shifted from broad ethics principles toward more operational taxonomies and implementation frameworks.	Earlier work (Jobin, Fjeld, Cath) concentrates on principles and broad governance challenges; later work (Hourglass, NIST, OECD accountability, Lu pattern catalogue, Batool SLR, Reuel technical governance) is more implementation- and taxonomy-oriented.	S1, S2, S3, S7, S8, S10, S11, S15, S17	High

4. Open disagreements / unresolved points

1. What counts as “AI governance”? Some sources use the term broadly to include ethics, standards, organizational process, and law (S1, S6, S15). Others narrow it to specific domains such as advanced/frontier AI (S12, S13) or technical governance questions (S17). The field still lacks a universally stable boundary.

2. How much weight should governance place on risk management versus rights and justice? Risk-based frameworks are highly influential (S10, S11), but rights-based and sociotechnical work argues that risk matrices can underweight power, exclusion, and democratic legitimacy (S1, S3, S6).

3. Is AI governance primarily an organizational design problem or a public-institution problem? Organizational models emphasize internal controls, policies, and lifecycle process (S5, S7, S8). Public-administration and international-policy literature emphasizes external oversight, legitimacy, public participation, and legal authority (S1, S4, S6).

4. How distinct is frontier/advanced AI governance from general responsible-AI governance? Advanced-governance literature treats highly capable systems as a distinct object requiring new concepts and institutions (S12, S13, S17). More general governance literature often presents lifecycle and accountability structures as broadly reusable across AI contexts (S6, S10, S11, S15).

5. How operational should governance frameworks be? Principles-oriented documents are easier to diffuse globally but weak on implementation (S2, S3). Pattern catalogues, RMFs, and organizational models are more actionable but may embed contested assumptions or under-specify politics and enforcement (S1, S8, S11).

6. Where should corporate governance sit within the AI-governance map? Some literature treats corporate structure as central to AI governance, especially where firms make public-purpose claims while controlling powerful systems (S16). Other literature still treats firm governance as secondary to regulation, standards, or ethics frameworks (S1, S15).

7. Can technical tools meaningfully solve governance problems? Technical AI governance work argues many governance questions are partly technical and need measurement, evaluations, interpretability, monitoring, and other tooling (S17). Sociotechnical critics warn against technical solutionism and over-compressing social values into metrics (S1).

5. Explicit ledger update note for T1

T1 status: completed. Artifact written to: outputs/ai-governance-research-literature.md