The DOI matches its Crossref record; the references it cites are not checked yet.
Potentials and Challenges of Large Language Models (LLMs) in the Context of Administrative Decision-Making
Open accessAbstract
Large Language Models (LLMs) could facilitate both more efficient administrative decision-making on the one hand, and better access to legal explanations and remedies to individuals concerned by administrative decisions on the other hand. However, it is an open research question of how performant such domain-specific models could be. Furthermore, they pose legal challenges, touching especially upon administrative law, fundamental rights, data protection law, AI regulation, and copyright law. The article provides an introduction into LLMs, outlines potential use cases for such models in the context of administrative decisions, and presents a non-exhaustive introduction to practical and legal challenges that require in-depth interdisciplinary research. A focus lies on open practical and legal challenges with respect to legal reasoning through LLMs. The article points out under which circumstances administrations can fulfil their duty to provide reasons with LLM-generated reasons. It highlights the importance of human oversight and the need to design LLM-based systems in a way that enables users such as administrative decision-makers to effectively oversee them. Furthermore, the article addresses the protection of training data and trade-offs with model performance, bias prevention and explainability to highlight the need for interdisciplinary research projects.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Their design also lacks intrinsic feedback loops [110], state management [111,112], or multi-step planning a requirement for autonomous decision-making and iterative goal refinement [113, 114].”
From AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges (2026) · cites this paper as [112]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 27
- Ilias Chalkidis et al., "LEGAL-BERT: The Muppets Straight Out of Law School" (2020) arXiv:2010.02559; · arXiv
- Shunyu Yao et al., "Lawyer GPT: A Legal Large Language Model with Enhanced Domain Knowledge and Reasoning Capabilities" in Proceedings of the 2024 3rd International Symposium on Robotics, Artificial Intelligence and Information Engineering (RAIIE '24) 108.
- See for example Jonathan J Choi and Daniel Schwarcz, "AI Assistance in Legal Analysis: An Empirical Study' 73 Journal of Legal Education (forthcoming 2024) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id= 4539836> accessed 10 December 2024;
- Lauren Martin et al., "Better Call GPT, Comparing Large Language Models Against Lawyers" (2024) arXiv:2401.16212; · arXiv
- John J Nay et al., "Large Language Models as Tax Attorneys: A Case Study in Legal Capabilities Emergence" 2024 (382) Philosophical Transactions of the Royal Society 20230159.
- Seungpill Lee et al., "Reasoning Abilities of Large Language Models: In-Depth Analysis on the Abstraction and Reasoning Corpus" (2024) arXiv:2403.11793. On Measuring Legal Reasoning in LLMs Neel Guha et al., "LEGALBENCH: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models" (2023) NeurIPS. Also cf Morison and McInerney (n 19) 26 f. · arXiv
- See for example Ilias Chalkidis, "ChatGPT may Pass the Bar Exam soon, but has a Long Way to go for the LexGLUE Benchmark" (2023) arXiv:2304.12202; · arXiv
- Wentao Deng et al., "Syllogistic Reasoning for Legal Judgment Analysis" in Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing 13997;
- Neel Guha et al., "LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models" (2023) arXiv:2308.11462; · arXiv
- Cong Jiang and Xiaolei Yang, "Legal Syllogism Prompting: Teaching Large Language Models for Legal Judgment Prediction" arXiv:2307.08321; · arXiv
- Xiaoxi Kang et al., "Can ChatGPT Perform Reasoning Using the IRAC Method in Analyzing Legal Scenarios Like a Lawyer?" in Findings of the Association for Computational Linguistics: EMNLP 2023, 13900 (Singapore, Association for Computational Linguistics);
- Daniel Martin Katz et al., "GPT-4 Passes the Bar Exam" (2024) 382 Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 20230254;
- Dietrich Trautmann, "Large Language Model Prompt Chaining for Long Legal Document Classification" (2023) arXiv:2308.04138; · arXiv
- Dietrich Trautmann, Alina Petrova and Frank Schilder, "Legal Prompt Engineering for Multilingual Legal Judgement Prediction" (2022) arXiv: 2212.02199; · arXiv
- Shaurya Vats, 'LLMs -the Good, the Bad or the Indispensable?: A Use Case on Legal Statute Prediction and Legal Judgment Prediction on Indian Court Cases" in Findings of the Association for Computational Linguistics: EMNLP 2023, 12451 (Singapore, Association for Computational Linguistics);
- Fangyi Yu, Lee Quartey and Frank Schilder, "Legal Prompting: Teaching a Language Model to Think Like a Lawyer" (2022) arXiv:2212.01326; · arXiv
- Fangyi Yu, Lee Quartey and Frank Schilder, "Exploring the Effectiveness of Prompt Engineering for Legal Reasoning Tasks" Findings of the Association for Computational Linguistics: ACL 2023, 1358 (Toronto, Canada, Association for Computational Linguistics). For a broader overview on legal argumentation based on natural language processing see Santosh T Y S S et al., "Towards Supporting Legal Argumentation with NLP: Is more Data Really All You Need?" (2024) arXiv:2406.10974v1 with further references. · arXiv
- Prompting" (2024) arXiv:2401.15585v1; · arXiv
- Roberto Navigli, Simone Conia and Björn Ross, "Biases in Large Language Models: Origins, Inventory, and Discussion" 2023 (15) Journal of Data and Information Quality no 10. On image generative models Mi Zhou et al., "Bias in Generative AI" (2024) arXiv:2403.02726. · arXiv
- Ganguli et al., "The Capacity for Moral Self-Correction in Large Language Models" (2023) arXiv:2302.07459v2. 91 See section II.1 above. 92 Ganguli et al., "The Capacity for Moral Self-Correction in Large Language Models" (2023) arXiv:2302.07459v2. 93 See sections II.3 and IV.7. 94 · arXiv
- Erfan Shayegani et al., "Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attacks" (2023) arXiv:2310.10844. · arXiv
- Jinwei Yi et al., "Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models" (2024) arXiv:2312.14197; · arXiv
- Yi Liu et al., "Prompt Injection Attack against LLM-Integrated Applications" (2024) arXiv:2306.05499v2. · arXiv
- On hallucination mitigation S M Towhidul Islam Tonmoy et al., "A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models" (2024) arXiv:2401.01313. · arXiv
- Finale Doshi-Velez et al., "Accountability of AI Under the Law: The Role of Explanation" revised (2019) arXiv: 1711.01134 14. 98 In the context of special categories of personal data according to Art 9 (1) GDPR CJEU Case C-184/20 Vyriausioji tarnybinės etikos komisija paras 117 ff. Also see Finale Doshi-Velez et al., "Accountability of AI Under the Law: The Role of Explanation" revised (2019) 14 · arXiv
- GDPR) comprises information that serves the norms' purpose and intent to enable the data subjects to exercise their individual rights. Also see Lilian Lilian Edwards and Michael Veale, "Slave To The Algorithm? Why A 'Right To An Explanation' Is Probably Not The Remedy You Are Looking For" 2017 (16) DLTR 18. 112 Parycek, Schmid and Novak (n 20) 8407 f; Hannah Ruschemeier, "The Problems of the Automation Bias in the Public Sector -A Legal Perspective" in Proceedings of the Weizenbaum Conference 2023: AI, Big Data, Social Media, and People on the Move, 1;
- Maia Jacobs et al., "How Machine-Learning Recommendations Influence Clinician Treatment Selection" 2021 Translational Psychiatry Article no. 8; Forough Poursabzi-Sangdeh et al., "Manipulating and Measuring Model Interpretability" (2021) arXiv:1802.07810. 122 Poursabzi-Sangdeh et al. (n 121). 123 Cf Marwa Gadala, "Automation Bias: Exploring Causal Mechanisms and Potential Mitigation Strategies" (2017) <https://openaccess.city.ac.uk/id/eprint/17889/> accessed 10 December 2024, 71. 124 In the context of the European Travel Information and Authorisation System (ETIAS) Pesch, Dimitrova and Boehm (n 107) 60. 125 Gadala (n 123) 35 ff. 126 On blockchain forensics Michael Fröwis et al., "Safeguarding the Evidential Value of Forensic Cryptocurrency Investigations" 2020 (33) FSI: Digital Investigation 200902. 127 In the context of LLM-based applicant selection Kätscher and Pesch (n 8) 51. · arXiv
Cited by
Sorted by citations
AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges
Sapkota, R., Roumeliotis, K. I., & Karkee, M. (2026) · Information Fusion · doi:10.1016/j.inffus.2025.103599 · Cited by 671
“Their design also lacks intrinsic feedback loops [110], state management [111,112], or multi-step planning a requirement for autonomous decision-making and iterative goal refinement [113, 114].”
AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges
Sapkota, R., Roumeliotis, K. I., & Karkee, M. (2025) · SuperIntelligence - Robotics - Safety & Alignment · doi:10.70777/si.v2i3.15161 · Cited by 186
Schulte, N., & Kanbach, D. K. (2026) · Management Review Quarterly · doi:10.1007/s11301-026-00611-2 · Cited by 16
Govea, J., Ortiz-Garcés, I., Palacios, P., Navarro, A. M., del Pino, S. M. A., & Villegas-Ch, W. E. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1782405 · Cited by 1
Marchetti, B., & Torchia, L. (2025) · European Union and its neighbours in a globalized world · doi:10.1007/978-3-032-06490-5_1 · Cited by 1
Metadata
- Authors
- Pesch, P. J.
- Affiliations
- Friedrich-Alexander-Universität Erlangen-Nürnberg (DE)
- Corresponding author
- Pesch, Paulina Jo
- ORCID
- Pesch, Paulina Jo
- Venue
- European Journal of Risk Regulation
- Publisher
- Cambridge University Press (CUP)
- Published
- 2025-01-31
- Volume
- 16
- Issue
- 1
- Pages
- 76-95
- ISSN
- 1867-299X (ISSN-L)
- 2190-8249
- OpenAlex
- W4407048454
- Cited by
- 12
- References
- 0
- Citations by year
- 2026: 10 · 2025: 2
- FWCI
- 44.80
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- European Journal of Risk Regulation (publishedVersion, cc-by-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Universität Innsbruck
- Topics
- Social Sciences › Law › Artificial Intelligence in Law
- Social Sciences › Law › Digitalization, Law, and Regulation
- Social Sciences › Law › Administrative Law and Governance
- Keywords
- large language models, administrative decisions, administrative law, duty to give reasons, legal reasoning, human oversight, legal explanation, remedies, fundamental rights, data protection law, AI regulation, copyright law, algorithmic bias, explainability
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.
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
