Administrative Law and the Machines of Government: Judicial Review of Automated Public-Sector Decision-Making
Open accessThe DOI matches its Crossref record; 8 of the 8 references it cites were found in a registry or an index; none of the 100 sources its claims cite is checked for support yet.
Abstract
The future is likely to see an increase in the public-sector use of automated decision-making systems which employ machine learning techniques. However, there is no clear understanding of how English administrative law will apply to this kind of decision-making. This paper seeks to address the problem by bringing together administrative law, data protection law, and a technical understanding of automated decision-making systems in order to identify some of the questions to ask and factors to consider when reviewing the use of these systems. Due to the relative novelty of automated decision-making in the public sector, this kind of study has not yet been undertaken elsewhere. As a result, this paper provides a starting point for judges, lawyers, and legal academics who wish to understand how to legally assess or review automated decision-making systems and identifies areas where further research is required.
Overview
CoChat Summary
## TL;DR
This paper argues that English administrative law can and should be applied to automated public-sector decision-making, but that doing so requires combining doctrinal review with data protection rules and technical understanding of machine learning systems. It identifies the main legal questions reviewers should ask and flags several areas where current law and tools are not yet adequate for meaningful oversight.
## Key findings
- Public bodies, not machines, remain legally responsible for decisions made by or with ADM, and ADM must meet the same administrative law standards as human decision-making.
- Where personal data is involved, GDPR heavily shapes legality: Article 22 prohibits solely automated decisions producing legal or similarly significant effects unless an exemption applies, and the paper notes that consent is usually unlikely to be valid in public-sector contexts because of the imbalance between individual and authority.
- The paper argues that review of ADM should examine not only the final decision but also training data, input data, and inferences/predictions: inaccuracies, irrelevance, or unnecessary processing can make a decision unlawful.
- It highlights practical review problems: the paper notes the current three-month judicial review time limit may be too short for claimants to obtain and assess the necessary data, and says extending it to 6, 9, or 12 months, or starting time when the data are received, could improve access to justice.
## Methods
- Doctrinal/legal analysis of English administrative law, GDPR, and the Data Protection Act 2018, combined with a technical discussion of how automated decision-making and machine learning systems work.
- The paper is framed as a high-level review of public-sector ADM rather than an in-depth analysis of sector-specific statutes or the detailed inner workings of ADM systems.
- It synthesizes legal grounds for judicial review with technical features of ADM to derive questions and factors for legal reviewers.
## Limitations
- The paper explicitly says it does not undertake an in-depth analysis of finer points of administrative law, sector-specific statutory requirements, or the intricacies of ADM systems.
- It states that due to the novelty of public-sector ADM, this kind of study has not yet been undertaken elsewhere, so the paper is a starting point rather than a settled framework.
- Evident from the text: many of the technical and legal review tools the paper calls for do not yet exist in widely used or easily accessible form, which limits immediate practical application.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 13 Not valid · 81 Not checked · 90 Own results and interpretations
In the paper
“increased transparency over the internal workings of models seems to reduce people's ability to detect even sizeable mistakes 16 .”
What the sources say
- Not checked yetcites [16] S Corbett-Davies, 2017, Algorithmic decision making and the cost of fairness source
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.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As a result, the processes and metrics for fair, accountable, and transparent machine learning developed through this research do not always translate easily to legal frameworks.”
hedged- 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.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The result of algorithmic opacity is that an automated system's decision -making process may be difficult to understand or impossible to evaluate even for experienced systems designers and engineers, let alone non-technical reviewers.”
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.
In the paper
“While researchers have sought to address this problem 14 , they have not yet succeeded to the extent that solutions -where available -are likely to be useful to a legal or otherwise non-technical audience.”
What the sources say
- Not checked yetcites [14] E G See, 2016, Inherent Trafe-Offs in the Fair Determination of Risk Scores
hedged- 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.
In the paper
“Machine learning systems are known to have various issues relating to bias, unfairness, and discrimination in decisions 9”
What the sources say
- Not checked yetcites [9] 2001, Re Medicaments and Related Classes of Goods
boostedstates 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.
98 more claims
Create a free account to see all 184 claims and to verify the 100 references still open.
- Probably not a substantial claim. 58
- It may state two separate claims as one. 39
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Use of ADM to exercise discretionary powers
“80 Australian Government, above n 19, p viii, p 37; see also Le Sueur, above n 3, pp 196-197”
§ Legal responsibility for ADM
“19 Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Use of contracted-out ADM
“105 Clayton, above n 110”
§ Training and decision data
“110 GDPR, art 5(1)(d)”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ The rule against bias
“128 Courtland, above n 156”
§ The rule against bias
“127 See, e.g., R Courtland”
Citations
Cites
Friedman, B., & Nissenbaum, H. (1996) · ACM Transactions on Information Systems · doi:10.1145/230538.230561 · Cited by 1,203
Algorithmic Decision Making and the Cost of Fairness
Corbett-Davies, S., Pierson, E., Feller, A., Goel, S., & Huq, A. (2017) · Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/3097983.3098095 · Cited by 1,149
Fairness in Criminal Justice Risk Assessments: The State of the Art
Berk, R., Heidari, H., Jabbari, S., Kearns, M., & Roth, A. (2021) · Sociological Methods & Research · doi:10.1177/0049124118782533 · Cited by 988
Bias detectives: the researchers striving to make algorithms fair
Courtland, R. (2018) · Nature · doi:10.1038/d41586-018-05469-3 · Cited by 205
Data protection as a fundamental right next to privacy? 'Reconstructing' a not so new right
Tzanou, M. (2013) · International Data Privacy Law · doi:10.1093/idpl/ipt004 · Cited by 72
Schröer, A. (2014) · Palgrave Macmillan eBooks · doi:10.1057/9781137326256.0015 · Cited by 36
Lifelong Privacy: Privacy and Identity Management for Life
Pfitzmann, A., & Borcea‐Pfitzmann, K. (2010) · IFIP International Federation for Information Processing/IFIP · doi:10.1007/978-3-642-14282-6_1 · Cited by 7
Anisminic Ltd v Foreign Compensation Commission [1968]: In Perspective
(2017) · Hart Publishing eBooks · doi:10.5040/9781474201902.ch-004 · Cited by 1
Research works printed but not linked: 7
- 2001, R (Nash) v Chelsea College of Art and Design
- R V Higher, Education Funding Council
- 1968, R v Minister of Agriculture Fisheries and Food, ex parte Padfield
- 2018, Factsheet -personal data protection, European Court of Human Rights
- 1996, R v Secretary of State for the Environment, ex parte Kirkstall Valley Campaign
- 2001, Re Medicaments and Related Classes of Goods
- E G See, 2016, Inherent Trafe-Offs in the Fair Determination of Risk Scores
Cited by
Alon-Barkat, S., & Busuioc, M. (2023) · Journal of Public Administration Research and Theory · doi:10.1093/jopart/muac007 · Cited by 469
Interpretable Machine Learning
Boehmke, B., & Greenwell, B. (2019) · Hands-On Machine Learning with R · doi:10.1201/9780367816377-16 · Cited by 230
Discrimination, artificial intelligence, and algorithmic decision-making
Borgesius, F. Z. (2025) · arXiv · doi:10.48550/arxiv.2510.13465 · Cited by 152
The Chinese Social Credit System: A Model for Other Countries?
Síthigh, D. M., & Siems, M. (2019) · Modern Law Review · doi:10.1111/1468-2230.12462 · Cited by 190
Human Judgment in algorithmic loops: Individual justice and automated decision‐making
Binns, R. (2022) · Regulation & Governance · doi:10.1111/rego.12358 · Cited by 134
Metadata
- Authors
- Cobbe, J.
- Affiliations
- University of Cambridge (GB)
- Corresponding author
- Cobbe, Jennifer
- ORCID
- Cobbe, Jennifer
- Venue
- SSRN Electronic Journal
- Publisher
- Elsevier BV
- Published
- 2018
- Volume
- 39
- Issue
- 4
- Pages
- 636-655
- ISSN
- 0261-3875 (ISSN-L)
- 1556-5068
- OpenAlex
- W3123074717
- Cited by
- 165
- References
- 8
- Citations by year
- 2026: 21 · 2025: 18 · 2024: 15 · 2023: 69 · 2022: 16 · 2021: 12 · 2020: 11 · 2019: 3
- FWCI
- 57.07
- Type
- journal article
- Language
- en
- Open access
- green
- doi.org
- Locations
- Legal Studies (publishedVersion)
- SSRN Electronic Journal (publishedVersion)
- Apollo (University of Cambridge) (submittedVersion)
- Apollo
- Funders
- Engineering and Physical Sciences Research Council
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Social Sciences › Law › Administrative Law and Governance
- Social Sciences › Law › Artificial Intelligence in Law
- Keywords
- automated decision-making, machine learning, public sector, administrative law, judicial review, data protection law, algorithmic governance, English administrative law
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.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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
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.