The DOI matches its Crossref record; 31 of the 31 references it cites were found in a registry or an index; 22 of the 67 sources its claims cite were checked for support.
Human-centric AI governance in the European Union. Accountability, fundamental rights, and institutional resilience in public administration
Open accessAbstract
Artificial intelligence is increasingly used in public administration to classify individuals, assess risks, prioritize cases, support eligibility determinations and guide the allocation of public resources. In the European Union, these uses are governed by the Artificial Intelligence Act, the GDPR and the Charter of Fundamental Rights. Formal compliance, however, does not by itself ensure lawful and accountable administration. AI relocates discretion from the visible act of decision-making to less visible choices concerning data, model design, procurement, thresholds and interface architecture. This Policy and Practice Review therefore treats human-centric AI governance not as a general ethical aspiration, but as an administrative and constitutional framework for governing public power. Drawing on EU law, public administration scholarship and a comparative institutional analysis of selected Member State practices, it develops six interdependent dimensions: legal anchoring, accountable discretion, fundamental rights by design, meaningful human oversight, contestability and justification, and institutional resilience. The analysis shows that common EU rules may produce unequal levels of protection where public authorities differ in technical expertise, audit capacity, procurement independence and access to effective remedies. It also argues that accountability must follow the chain of influence through which algorithmic systems shape administrative outcomes, rather than only the formal chain of decision-making. The article translates this framework into actor-specific recommendations concerning fundamental rights impact assessments, procurement, auditability, human oversight, transparency, contestability and post-deployment monitoring. It concludes that AI-enabled administration remains legitimate only where public authorities retain the capacity to understand, justify, correct, suspend and democratically control the systems they use.
Overview
CoChat Summary
TL;DR
This Policy and Practice Review argues that human-centric AI governance in EU public administration must be treated as an administrative and constitutional framework for governing public power, not a general ethical aspiration. It develops a six-dimensional framework and shows, through a comparative reading of the Netherlands, France and Estonia, that harmonized EU rules can produce unequal protection where authorities differ in expertise, audit capacity, procurement independence and access to remedies.
Key findings
- AI does not eliminate administrative discretion but relocates it upstream into data selection, model design, threshold setting, procurement and interface architecture; accountability must therefore follow the "chain of influence" rather than only the formal chain of decision-making.
- The article proposes six interdependent dimensions: legal anchoring, accountable discretion, fundamental rights by design, meaningful human oversight, contestability and justification, and institutional resilience (Table 2), each tied to distinct legal anchors (AI Act, GDPR, Charter Articles 41 and 47, NIS2 where applicable) and institutional safeguards.
- The comparative analysis locates safeguards at different stages: the Netherlands relies on ex post judicial correction (SyRI, Rechtbank Den Haag 2020, finding the legislation failed the fair balance under Article 8(2) ECHR); France embeds statutory procedural transparency in administrative law (Articles L311-3-1 and R311-3-1-1 of the Code des relations entre le public et l'administration); Estonia illustrates coordinated digital-government capacity and shared infrastructure (OECD 2026a), which the authors stress is an enabling condition rather than a substitute for rights safeguards.
- The paper argues that formal compliance with the AI Act's risk-based architecture (Articles 9–15, 14, 27, 71, 86) and Article 22 GDPR cannot by itself secure legitimate administrative use, because risk is both technological and institutional; nominal human involvement may reduce the apparent relevance of Article 22 GDPR even where algorithmic output retains decisive influence (citing Sarra, 2025).
Methods
- Design: structured legal-doctrinal and comparative governance approach; the article explicitly does not measure AI adoption levels or assess technical performance of individual systems.
- Corpus: primary legal texts (AI Act, GDPR, Charter of Fundamental Rights, relevant national administrative law and judicial decisions, with NIS2 considered only for the institutional-resilience dimension) plus policy documents (2019 Ethics Guidelines for Trustworthy AI, EU implementation materials) and secondary literature identified via targeted searches in Scopus, Web of Science, HeinOnline and Google Scholar, with citation tracing, covering materials up to 20 July 2026.
- Comparative component: purposive, maximum-variation selection of the Netherlands, France and Estonia, examined against a common grid (legal and procedural safeguards, location of oversight, administrative and technical capacity, provider dependence, transparency and contestability, capacity for correction and learning); analysis proceeds through four operations — identifying legal duties, tracing discretion across the process, examining the three settings against the grid, and cross-case comparison feeding the six-dimensional framework.
Limitations
- Stated in the paper (Section 12): the study relies on legal, policy and institutional documents and does not assess technical performance or measurable effects of individual AI systems; the three national settings were selected for contrasting features and do not support a general ranking of Member States; documentary evidence varies across jurisdictions and several AI Act mechanisms remain in development; the framework supports qualitative institutional inference rather than causal generalization, and its practical value should be tested through further empirical research on procurement, FRIAs, human oversight, administrative remedies and post-deployment monitoring.
- Evident from the text: the comparative cases are not treated as representative of all algorithmic use within their administrations, and the authors note that SyRI's precise technical characteristics were not fully public and the judgment did not establish a general prohibition on automated fraud detection (citing Van Bekkum and Zuiderveen Borgesius, 2021).
- Evident from the text: the article is a Policy and Practice Review whose recommendations are derived from documentary and comparative inference; no primary empirical data (e.g., interviews, procurement records, audit results) are collected or analyzed.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
22 Valid · 3 Not valid · 43 Not checked · 176 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“High-risk AI systems are socio-technical arrangements in which fairness, privacy, accuracy and operational efficiency may come into tension (Sunyaev et al., 2025).”
What the sources say
- Supportedcites Sunyaev et al., 2025 source
Section 2.1 (p. 3) describes a tradeoff between predictive accuracy and fairness. Section 2.2 (p. 4) links privacy-preserving approaches to potential reductions in accuracy and operational efficiency, while Section 5 (p. 9) says these dilemmas require socio-technical approaches. (LLM, conf=0.93)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Yet the insertion of a human actor may reduce the apparent relevance of Article 22 GDPR even where the algorithmic output continues to exert decisive practical influence (Sarra, 2025).”
What the sources say
- Supportedcites Sarra, 2025 sourceExperimental — accuracy not yet measured
In “The Extra Safeguard Measures” (p. 6), Sarra explains that under the prevailing interpretation of Article 22, a decision may be treated as not fully automated because a qualified human is present, even if that person merely rubber-stamps a decision made by the machine. “A Different Interpretation” (p. 11) says this can make Article 22 inapplicable; the paper argues for changing that interpretation. (LLM, conf=0.94)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“As Sabel and Zeitlin show, complex governance problems are often addressed through iterative rules and institutional learning rather than through one-off commands (Sabel and Zeitlin, 2008).”
What the sources say
- Supportedcites Sabel and Zeitlin, 2008 source
The Introduction describes recursive rule-making in which EU frameworks are adapted to local circumstances and revised through experience. The discussion of experimentalist governance (page not supplied) contrasts this learning process with command-and-control administration. This supports the citing sentence’s paraphrase. (LLM, conf=0.91)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“The subsequent move from ethical guidance to binding regulation formed part of a broader shift from a 'race to AI' toward a race to AI regulation' (Smuha, 2021).”
What the sources say
- Supportedcites Smuha, 2021 source
The Introduction says the ‘race to AI’ is bringing forth a ‘race to AI regulation’. Section 4 describes the EU’s announced move from voluntary AI ethics guidelines toward binding legislation, matching the claimed transition; the paper presents binding legislation as prospective, not already enacted. (LLM, conf=0.90)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Risk-based regulation remains appropriate where it is proportionate and evidence-based (Ebers, 2025)”
What the sources say
- Supportedcites Ebers, 2025 source
In “Essential features of risk-based regulation,” Ebers says risk assessment requires sufficient empirical evidence and that regulatory burdens should be proportionate to the risks posed. The same section calls risk-based regulation the right approach to AI regulation when properly implemented, while acknowledging its limitations. (LLM, conf=0.93)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Supported
119 more claims · 6 with their cited sources checked
Create a free account to see all 244 claims and to verify the 2 references still open.
- It may state two separate claims as one. 83
- Probably not a substantial claim. 68
- The quote only partly states this claim. 2
Citations
Cites
Multi-Level Governance and European Integration
Hooghe, L., & Marks, G. (2002) · Rowman & Littlefield Publishers eBooks · doi:10.5040/9798216401537 · Cited by 2,598
Bainbridge, L. (1983) · Automatica · doi:10.1016/0005-1098(83)90046-8 · Cited by 2,205
Principles alone cannot guarantee ethical AI
Mittelstadt, B. (2019) · Nature Machine Intelligence · doi:10.1038/s42256-019-0114-4 · Cited by 1,886
van Kersbergen, K., & van Waarden, F. (2004) · European Journal of Political Research · doi:10.1111/j.1475-6765.2004.00149.x · Cited by 971
Bovens, M., & Zouridis, S. (2002) · Public Administration Review · doi:10.1111/0033-3352.00168 · Cited by 964
Veale, M., & Zuiderveen Borgesius, F. (2021) · Computer Law Review International · doi:10.9785/cri-2021-220402 · Cited by 805
Learning from difference: the new architecture of experimentalist governance in the EU
Sabel, C. F., & Zeitlin, J. (2008) · European Law Journal · doi:10.1111/j.1468-0386.2008.00415.x · Cited by 791
Accountable Artificial Intelligence: Holding Algorithms to Account
Busuioc, M. (2021) · Public Administration Review · doi:10.1111/puar.13293 · Cited by 652
Kuziemski, M., & Misuraca, G. (2020) · Telecommunications Policy · doi:10.1016/j.telpol.2020.101976 · Cited by 610
Unraveling the Central State, but How? Types of Multi-level Governance
LIESBET, H., & GARY, M. (2003) · American Political Science Review · doi:10.1017/s0003055403000649 · Cited by 489
Research works printed but not linked: 5
- 2012, Charter of fundamental rights of the European Union, Off. J. Eur. Union C, 326, 391
- 2016, Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation), Off. J. Eur. Union L, 119, 1
- 2022, Directive (EU) 2022/2555 of the European parliament and of the council of 14 December 2022 on measures for a high common level of cybersecurity across the Union (NIS 2 Directive), Off. J. Eur. Union L, 333, 80
- 2024, Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), Off. J. Eur. Union L
- Digital Government Outlook 2026: From Foundations to Transformational Impact
Cited by
Rexhepi, I., & Islami, X. (2026) · The Bottom Line Managing Library Finances · doi:10.1108/bl-03-2026-0083
Metadata
- Authors
- Dragomir, A. N., Bulea, I., & Tarnu, L. I.
- Affiliations
- Lucian Blaga University of Sibiu (RO)
- Corresponding author
- Dragomir, Andreea Nicoleta
- Venue
- Frontiers in Political Science
- Publisher
- Frontiers Media SA
- Published
- 2026-08-14
- Volume
- 8
- ISSN
- 2673-3145 (ISSN-L)
- OpenAlex
- W7203459260
- Cited by
- 1
- References
- 30
- Citations by year
- 2026: 1
- FWCI
- 34.24
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Political Science (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Law › Administrative Law and Governance
- Keywords
- AI governance, public administration, European Union, EU AI Act, GDPR, Charter of Fundamental Rights, accountability, meaningful human control, algorithmic decision-making, administrative discretion, contestability, institutional resilience, fundamental rights impact assessment, auditability, transparency, post-deployment monitoring, effective remedy, EU member states
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Human-centric AI governance in the European Union. Accountability, fundamental rights, and institutional resilience in public administration. https://app.cochat.ai/papers/10.3389/fpos.2026.1928481
- The paper
- Dragomir, A. N., Bulea, I., & Tarnu, L. I. (2026). Human-centric AI governance in the European Union. Accountability, fundamental rights, and institutional resilience in public administration. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1928481
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.