Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models
Open accessThe DOI matches its Crossref record; 157 of the 158 references it cites were found in a registry or an index, and 1 are not checked yet; 29 of the 52 sources its claims cite were checked for support.
Abstract
Abstract Do large language models (LLMs) know the law? LLMs are increasingly being used to augment legal practice, education, and research, yet their revolutionary potential is threatened by the presence of “hallucinations”—textual output that is not consistent with legal facts. We present the first systematic evidence of these hallucinations in public-facing LLMs, documenting trends across jurisdictions, courts, time periods, and cases. Using OpenAI’s ChatGPT 4 and other public models, we show that LLMs hallucinate at least 58% of the time, struggle to predict their own hallucinations, and often uncritically accept users’ incorrect legal assumptions. We conclude by cautioning against the rapid and unsupervised integration of popular LLMs into legal tasks, and we develop a typology of legal hallucinations to guide future research in this area.
Verified 2026-11-19 · 1 reference not checked yet
Overview
CoChat Summary
TL;DR
This paper provides the first systematic evidence that public LLMs frequently hallucinate on verifiable questions about U.S. federal case law, with hallucination rates ranging from 58% for ChatGPT 4 to 88% for Llama 2. It also documents correlates of hallucination and two additional failure modes: susceptibility to false legal premises and poor calibration about their own accuracy.
Key findings
- Hallucinations are widespread across all tested LLMs on direct, verifiable federal case questions: ChatGPT 4 hallucinates 58% of the time, while Llama 2 hallucinates 88%; GPT 4 performs best overall, followed by GPT 3.5, PaLM 2, and Llama 2.
- Hallucination rates increase with task complexity and vary by court hierarchy, jurisdiction, case prominence, and case year. For reference-free high-complexity tasks, the reported rates are lower bounds: at least 49% for factual background and procedural posture, at least 59% for core legal question, and at least 63% for central holding; doctrinal agreement is near 0.5, little better than random guessing.
- LLMs often accept false legal premises: on the false dissent premise, GPT 4 hallucinates 0.691 at SCOTUS and 0.842 at USCOA; on the false overruling premise, GPT 4 hallucinates 0.531, GPT 3.5 0.821, PaLM 2 1.000, and Llama 2 0.027. Llama 2’s low contra-factual rate reflects its tendency to reject premises, which harms its performance on other tasks such as Existence.
- Calibration is poor and systematically overconfident: pooled expected calibration error is 0.057 for PaLM 2, 0.099 for GPT 3.5, 0.190 for GPT 4, and 0.421 for Llama 2. Although GPT 4 has the lowest raw hallucination rate, it is less calibrated than PaLM 2 and GPT 3.5.
Methods
- The authors develop 14 legal knowledge QA tasks grouped into low, moderate, and high complexity. The first nine are reference-based and scored against metadata from the Caselaw Access Project, Supreme Court Database, Appeals Courts Database Project, Library of Congress, and Shepard’s Citations; the final five are reference-free and detected via self-contradiction between repeated stochastic responses, using GPT 4 for contradiction labeling.
- They draw 5,000 cases each from SCOTUS, USCOA, and USDC using stratified random sampling, then query four off-the-shelf LLMs—OpenAI ChatGPT 4, OpenAI ChatGPT 3.5, Google PaLM 2, and Meta Llama 2—under zero-shot and three-shot prompting. They execute more than 800,000 queries, over 200,000 per LLM, generally at temperature 0 for reference-based tasks.
- Analyses estimate population-level hallucination rates, heterogeneity across task, court, jurisdiction, prominence, and year, plus contra-factual bias and expected calibration error.
Limitations
- The paper states that reference-free hallucination rates are only lower bounds because self-contradiction guarantees hallucination but logically consonant answers may still be false; it also cannot capture repeated incorrect answers as hallucinations, and notes that the SCOTUS Subsequent history estimate likely understates hallucination by a larger margin.
- The paper states that it relies on GPT 4 for contradiction detection, though it validates this with human intercoder reliability (Cohen’s κ: Coder 1 vs. GPT 4 = 0.87; Coder 2 vs. GPT 4 = 0.77; Coder 1 vs. Coder 2 = 0.79). It also notes that closed-source models do not expose conditional probabilities, so calibration is estimated by sampling at temperature 1, and that ex post temperature scaling improves but does not fully solve calibration.
- Evident from the text: the study tests specific public model versions available at the time, so results may not generalize to newer LLMs or to retrieval-augmented, fine-tuned, or proprietary legal systems; the tasks are limited to published federal case law and to metadata available for constructing ground truth.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
16 Valid · 5 Not valid · 9 Not checked · 105 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“since hallucinations of some kind are generally inevitable at the token level (Kalai and Vempala 2023; Xu, Jain, and Kankanhalli 2024)”
What the sources say
- Disputedcites Kalai and Vempala 2023 matched to “Calibrated Language Models Must Hallucinate” (2023)Experimental — accuracy not yet measured
The Introduction (p. 1) explicitly says the paper's lower bounds for calibrated predictors of facts should not be taken to mean hallucination is inevitable. The discussion on p. 2 distinguishes its semantic-level calibration from token-level calibration, so it does not establish inevitability at the token level. (LLM, conf=0.87)
Checked by a reasoning model — may be wrong.
- Partly supportedcites Xu, Jain, and Kankanhalli 2024 matched to “Hallucination is Inevitable: An Innate Limitation of Large Language Models” (2024) sourceExperimental — accuracy not yet measured
Section 3.2 (p. 7) concludes that LLM hallucination cannot be eliminated, supporting the inevitability claim. But Section 2.1 (p. 2) describes token-by-token generation, while the paper defines hallucination by comparing complete output strings with ground truth; it does not establish inevitability specifically at the token level. (LLM, conf=0.85)
plain- 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.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“as Kalai and Vempala (2023) show, LMs that assign a positive probability to every response token must hallucinate at least some of the time”
What the sources say
- Supportedcites Kalai and Vempala (2023) matched to “Calibrated Language Models Must Hallucinate” (2023)
The Introduction (p. 1) explicitly says that an LM predicting every string with positive probability will necessarily hallucinate with positive probability. This supports the claim's paraphrase that such hallucination occurs at least some of the time. (LLM, conf=0.93)
boostedstates a 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
“Xu, Jain, and Kankanhalli (2024) agree, arguing that 'hallucination is inevitable for any computable LLM, regardless of model architecture, learning algorithms, prompting techniques, or training data.'”
What the sources say
- Supportedcites Xu, Jain, and Kankanhalli (2024) matched to “Hallucination is Inevitable: An Innate Limitation of Large Language Models” (2024) source
The Introduction (p. 2) states the attributed claim nearly verbatim. It frames the result under the paper’s formal definition of hallucination and says it applies to real-world LLMs. (LLM, conf=0.95)
plain- 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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“increased reliance on LLMs also has the potential to produce a kind of legal monoculture (Creel and Hellman 2022; Kleinberg and Raghavan 2021), with users being fed information from only a limited subset of judicial sources that elide many of the deeper nuances of the law.”
What the sources say
- Partly supportedcites Creel and Hellman 2022 matched to “The Algorithmic Leviathan: Arbitrariness, Fairness, and Opportunity in Algorithmic Decision-Making Systems” (2022) sourceExperimental — accuracy not yet measured
Section 3.a, “Why arbitrariness at scale arises” (page unavailable), says that reliance on similar datasets can standardize different algorithms and produce uniform outcomes. That supports the general concern about algorithmic monoculture, but the paper does not address LLMs feeding users a limited subset of judicial sources or eliding nuances of law. (LLM, conf=0.84)
- Partly supportedcites Kleinberg and Raghavan 2021 matched to “Algorithmic monoculture and social welfare” (2021) sourceExperimental — accuracy not yet measured
The Introduction (p. 1) describes algorithmic curation making choices more homogeneous, and the Conclusion (p. 12) says shared evaluations can cause valuable options to be overlooked. Those passages support the general possibility of monoculture, but the paper examines algorithmic screening, not LLMs or whether users receive a limited subset of judicial sources that omits legal nuance. (LLM, conf=0.87)
hedgedstates a possible cause- 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.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“the problem of legal hallucination has so far only been studied in closed-domain applications, such as when a model is used to summarize the content of a given judicial opinion (Deroy, Ghosh, and Ghosh 2023; Feijo and Moreira 2023) or to synthesize provided legal text (Savelka et al. 2023)”
What the sources say
- Partly supportedcites Deroy, Ghosh, and Ghosh 2023 matched to “How Ready are Pre-trained Abstractive Models and LLMs for Legal Case Judgement Summarization?” (2023) sourceExperimental — accuracy not yet measured
The Abstract describes a study of hallucinations in summaries generated from Indian court case judgments, supporting this paper as an example of a closed-domain legal application. The Introduction’s “Novelty of this work” section says that, to the authors’ knowledge, hallucination had not previously been studied in legal summarization. Neither passage establishes the broader claim that legal hallucination had only been studied in closed-domain applications. (LLM, conf=0.84)
- Not checked — no full text of the sourcecites Feijo and Moreira 2023 matched to “Improving Abstractive Summarization of Legal Rulings through Textual Entailment” (2023) source
- Partly supportedcites Savelka et al. 2023 matched to “Explaining Legal Concepts with Augmented Large Language Models (GPT-4)” (2023) sourceExperimental — accuracy not yet measured
The Proposed Method (pp. 6–7) describes an augmented setup that synthesizes retrieved case-law sentences into an explanation, supporting the example attributed to Savelka et al. But it also studies a baseline that generates explanations without those retrieved sentences, and the Conclusions (p. 13) report hallucinated case-law citations in that baseline. The paper therefore does not support the claim that legal hallucination had only been studied in closed-domain applications. (LLM, conf=0.87)
plain- 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.
- Partly supported
60 more claims · 3 with their cited sources checked
Create a free account to see all 135 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 42
- It may state two separate claims as one. 44
- The quote only partly states this claim. 1
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
“In high-stake domains such as legal consultation or scientific research, these hallucinations can lead to severe misjudgments and erode user trust [190,191].”
From AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges (2026) · cites this paper as [190]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Recent research has shown that leading language models – ChatGPT, Deepseek, Grok, PaLM, Claude, Qwen, and Llama – were on average producing hallucinations in 75 % of the cases (Dahl et al., 2024), with many users unaware of this prevalence (Williamson & Prybutok, 2024).”
From If ChatGPT can do it, where is my creativity? generative AI boosts performance but diminishes experience in creative writing (2025) · cites this paper as Dahl et al., 2024
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Dahl et al. [7] showed that even models with high bar exam scores systematically fabricate case references when applied to real legal tasks.”
From Citation Grounding: Detecting and Reducing LLM Citation Hallucinations via Legal Citation Graphs (2026) · cites this paper as [7]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“a confident legal summary can make uncertain advice feel authoritative [93]”
From Caring Without Feeling: Affective Dynamics as the Control Layer of Human-AI Agent Collaboration (2026) · cites this paper as [93]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Previous work has found that general- purpose LLMs hallucinate on legal queries on average between 58% and 82% of the time (Dahl et al. 2024).”
From Hallucination‐Free? Assessing the Reliability of Leading AI Legal Research Tools (2025) · cites this paper as Dahl et al. 2024
Experimental — accuracy not yet measured
10 more claims about this paper
Create a free account to see every claim other papers make about this one.
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
§ E.1 Fake Case Existence Task · p. 61
“GPT 4 and Llama 2, on the other hand, appear immune from this behavior, but recall from Table 2 that this is just bias in the opposite direction: they simply prefer to deny the existence of any case, real or fake.”
§ 6 Discussion · p. 38
“| Existence | Zero-shot | 0.204 (0.006) | 0.004 (0.001) | 0.054 (0.003) | 0.303 (0.006)”
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
§ 5.1.2 Hallucinations Vary by Court · p. 24
“We find that across all tasks and all LLMs, hallucinations are lowest in the highest levels of the judiciary, and vice-versa (Figure 3).”
§ 6 Discussion · p. 38
“| Citation | Few-shot | 0.689 (0.007) | 0.641 (0.007) | 0.833 (0.005) | 0.953 (0.003) | 0.468 (0.007)”
Citations
Cites
The Measurement of Observer Agreement for Categorical Data
Landis, J. R., & Koch, G. G. (1977) · Biometrics · doi:10.2307/2529310 · Cited by 81,148
A Coefficient of Agreement for Nominal Scales
Cohen, J. (1960) · Educational and Psychological Measurement · doi:10.1177/001316446002000104 · Cited by 42,398
Tversky, A., & Kahneman, D. (1974) · Science · doi:10.1126/science.185.4157.1124 · Cited by 28,810 · cited 1×
“…(the tendency of an LLM to agree with a user’s preferences or beliefs, even when the LLM would reject the belief as wrong without the user’s prompting; Sharma et al., 2023; Wei et al., 2023) or general cognitive error (Tversky and Kahneman, 1974; Jones and Steinhardt, 2022; Suri et al., 2023).”
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
Kaugeranna, R., Kaugeranna, E., & Aion, (. S. 4. 6. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.giscience.2023.43 · Cited by 13,690
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025
On calibration of modern neural networks
Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017) · arXiv · doi:10.48550/arxiv.1706.04599 · Cited by 3,435
Barocas, S., & Selbst, A. D. (2014) · doi:10.2139/ssrn.2477899 · Cited by 3,048
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
Self-citationLlama 2: Open Foundation and Fine-Tuned Chat Models
Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.09288 · Cited by 2,604
Research works printed but not linked: 10
- Ambrogi, Bob. 2023. 'As Allen & Overy Deploys GPT-based Legal App Harvey Firmwide, Founders Say Other Firms Will Soon Follow.' LawSites.
- Chien, Colleen V., Miriam Kim, Raj Akhil, and Rohit Rathish. 2024. 'How Generative AI Can Help Address the Access to Justice Gap Through the Courts.' Loyola of Los Angeles Law Review, forthcoming.
- Draper, Chris, and Nicky Gillibrand. 2023. 'The Potential for Jurisdictional Challenges to AI or LLM Training Datasets.' In Proceedings of the ICAIL 2023 Workshop on Artificial Intelligence for Access to Justice. Braga, Portugal: CEUR Workshop Proceedings.
- Dworkin, Ronald. 1986. Law's Empire. Cambridge, MA: Harvard University Press.
- Luft, Joseph, and Harrington Ingham. 1955. 'The Johari Window as a Graphic Model of Interpersonal Awareness.' In Proceedings of the Western Training Laboratory in Group Development. University of California, Los Angeles, Extension Office.
- Mayson, Sandra G. 2019. 'Bias In, Bias Out.' The Yale Law Journal 128 (8): 2122-2473.
- Romoser, James. 2023. 'No, Ruth Bader Ginsburg Did Not Dissent in Obergefell - and Other Things ChatGPT Gets Wrong about the Supreme Court.' SCOTUSblog.
- Simshaw, Drew. 2022. 'Access to A.I. Justice: Avoiding an Inequitable Two-Tiered System of Legal Services.' Yale Journal of Law & Technology 24:150-226.
- Tan, Jinzhe, Hannes Westermann, and Karim Benyekhlef. 2023. 'ChatGPT as an Artificial Lawyer?' In Proceedings of the ICAIL 2023 Workshop on Artificial Intelligence for Access to Justice. Braga, Portugal: CEUR Workshop Proceedings.
- Weiser, Benjamin. 2023. 'Here's What Happens When Your Lawyer Uses ChatGPT.' The New York Times.
Cited by
Items 201–300 of 306
Soft-Community Kernel Rényi Spectrum for Semantic Uncertainty Estimation in Large Language Models
Li, Z., & Du, J. (2026) · Entropy · doi:10.3390/e28040442
“…risk interpretation [5]; in legal or financial assistance, semantically distinct answers may imply different obligations, recommendations, or decisions [6]; and in retrieval-augmented generation, a model may generate responses that sound confident despite deviating from the retrieved evidence [7].”
The Evolving Role of Large Language Models in Health Care Research and Scientific Communication
Zand, J., Šuvakov, M., Overgaard, S. M., Silvano, C. J., Meiners, L. M., Fan, J. W., Carter, R. E., Wieland, M. L., Patten, C. A., Dugani, S. B., et al. (2026) · Mayo Clinic Proceedings · doi:10.1016/j.mayocp.2026.04.007
Tang, L. (2026) · International Symposium on Big Data and Applied Statistics (ISBDAS) · doi:10.1109/isbdas69350.2026.11485008 · Cited by 1
“[8][9][10][11] Zheng et al. [12] proposed a lightweight LJP method tailored for resource-constrained devices in IoT environments based on previous research.”
Schweitzer, S., & Conrads, M. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6650978
Huang, Y., Chen, L., Han, S., & Guo, Y. (2026) · Research Square · doi:10.21203/rs.3.rs-8985237/v1
Zhang, Y., Wang, Y., Chen, S., et al. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6669386
Multi-Agent Orchestration of Local LLMs for Contract Review under Resource Constraints
An, Y., Wu, Z., Yang, Z., Zhou, S., Yin, Y., Wang, X., & Qin, P. (2025) · International Conference on Big Data, Artificial Intelligence and Risk Management (ICBAR) · doi:10.1145/3800227.3800266
Žalnieriūtė, M., & Limantė, A. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009744225.036
Junoy, J. P. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009744225.042
Žalnieriūtė, M., & Limantė, A. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009744225.003
The History and Development of AI in Courts: A Legal Informatics Perspective
Contissa, G., Galli, F., & Sartor, G. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009744225.005
The Cost Floor Theorem Why AI Answers Remain ExpensivePreprint
Dogonowski, R. R. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6717018 · Cited by 1
Hastings, C., Cotterell, A., & Bruce, F. (2026) · Australian Journal of Social Issues · doi:10.1002/ajs4.70115
‘Polyanna policy’ – is NZ’s framework for AI use in government overly optimistic?
Kawa, D. T., & Allen, B. (2026) · doi:10.64628/aa.egu4ak9tq
Liu, Q. (2026) · International Journal of Digital Law and Governance · doi:10.1515/ijdlg-2025-0021
“This is especially critical because hallucination is increasingly recognized not as a correctable anomaly but as an inherent feature of probabilistic language modeling (Dahl et al. 2024), making user awareness and independent veri fi cation indispensable.”
Schmitz, A. J., Giacalone, M., & Ortolani, P. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009589086.027
QUANTIFYING FINANCIAL AND REPUTATIONAL LIABILITY OF AI HALLUCINATIONS IN LEGAL PRACTICE
ДЕМІДОНТ, Б. (2026) · Підприємництво та інновації · doi:10.32782/2415-3583/39.33
ReGA: Zero-Overhead Graph Alignment for Structural Hallucination Detection Without Generation
Noël, V., & Seidou, E. Y. (2026) · International World Wide Web Conference (WWW) · doi:10.1145/3774905.3794657
“Recentwork onlegalAIhashighlightedtheneedforentity-levelprecision[7].”
Schneider, J. (2026) · Liverpool Law Review · doi:10.1007/s10991-026-09413-8
“Dahl and his co-authors show even more directly that apparently authoritative legal outputs can be fabricated with alarming frequency (Dahl et al. 2024, 64–65, 87).”
Wankhade, A., & Shahade, A. K. (2026) · The International journal of networked and distributed computing · doi:10.1007/s44227-026-00105-3
TCK-QA: Benchmarking Knowledge Graph-Augmented LLMs for Question Answering on the Turkish Penal CodePreprint
Ay, B., & Okuyan, H. (2026) · Research Square · doi:10.21203/rs.3.rs-9804227/v1
Automatic extraction and collection of legal principles from CJEU decisions on Fiscal State Aid
Perçin, S., Liepiņa, R., Galassi, A., Lagioia, F., Ruggeri, F., Santin, P., Sartor, G., & Torroni, P. (2026) · Artificial Intelligence and Law · doi:10.1007/s10506-026-09526-5
Comito, C. G., & Badessi, F. (2026) · Beijing Law Review · doi:10.4236/blr.2026.172032
Warren, G., Sun, J., Shklovski, I., & Augenstein, I. (2026) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3805689.3812358 · Cited by 4
“However, these artificial intelligence (AI) models suffer from issues with factuality [4, 12, 27] and bias [20, 44] in their outputs, which means that relying on LLMs for decision-making is risky.”
Zouak, F., Beqqali, O. E., & Riffi, J. (2026) · Big Data and Cognitive Computing · doi:10.3390/bdcc10070212
Auditing LLM Responses in a Complex Policy Landscape: Abortion Law in the United States
Encarnación, R., Jones, C., & Metaxa, D. (2026) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3805689.3806451
Vasudevan, I., & Gupta, S. (2026) · Lecture notes in networks and systems · doi:10.1007/978-981-95-7238-0_33
Confirmation Bias in Large Language Models: Risks for Clinical Decision Support Systems
Jardim, R. R.-A. J., Gonçalves, G., Lopes, L. S., Dantas, T., Paulino, D., Netto, A. T. C., Guimarães, D., Teixeira, O., Rocha, A., Vivacqua, A. S., et al. (2026) · International Conference on Computer Supported Cooperative Work in Design (CSCWD) · doi:10.1109/cscwd68734.2026.11581664 · Cited by 1
“[12] highlights the seriousness of hallucinations, particularly in contexts where fabricated information can have significant consequences.”
Elkafrawy, P., Hakim, A., Balubaid, A., & Alkhalili, F. (2026) · Learning and Technology Conference (L&T) · doi:10.1109/lt68265.2026.11592520
“Dahl et al. [1], [23] reported that hallucination rates are around 58% - 80% for general LLMs on legal tasks.”
Technologies, Applications, and Challenges of Large Language Models
Ahmad, I., Amelio, A., Aracne, M., Caroprese, L., Gill, E. Z., & Christian Morbidoni, B. (2026) · International Convention on Information and Communication Technology, Electronics and Microelectronics · doi:10.1109/mipro70003.2026.11591567 · Cited by 1
Navigating Examination Regulations: Reliable Chatbot Support for Students
Reid, S. A., Kammer, F., Lang, N., Ritz, H., Siepermann, M., & Wölfer, J. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-30784-2_30
PromptED: An Interactive Tool for Generative AI Literacy for Media Professionals
Novak, J., Tang, A., & Sowder, N. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-30784-2_15
Ali, G. K., & Jarrah, S. A. (2026) · IGI Global eBooks · doi:10.4018/979-8-3373-9179-3.ch007
Chen, Y., Zeng, Z., & Hassan, M. S. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1784200
Kennedy-Mayo, D., & Gord, J. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-24804-6_5
Zhou, M., Liu, T., Li, M., et al. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-3432-5_6
Chung, J. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7095298
Retrieval-augmented generation systems for intellectual property
Ren, R., Ma, J., & Luo, J. (2026) · Computers in Industry · doi:10.1016/j.compind.2026.104518 · Cited by 1
Uncertainty Quantification for Multimodal Retrieval Augmented Generation
Binz, S., Soudani, H., & Hasibi, F. (2026) · ACM SIGIR International Conference on Theory of Information Retrieval (ICTIR) · doi:10.1145/3805713.3820431 · Cited by 2
“This is particularly critical in domains such as medical or legal applications, where incorrect responses can have severe consequences [3, 8, 15].”
VCL: Bridging Natural Language and Structured Control for Domain-Specific Document Analysis
Principe, R. A. A., Armani, F., Palmonari, M., & Batini, C. (2026) · International Workshop on Exploiting Artificial Intelligence Techniques for Data Management (aiDM) · doi:10.1145/3814940.3815330
Liu, L., Yao, S., Liu, Y. P., Wang, Y., Xia, H., & Tang, X. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.105027
Bankura, S., Kalaivani, J., Veeraputhiran, A., & Rama, P. (2026) · IEEE Nexo-Tech International Conference on Advanced Technologies and Innovations (ICATI) · doi:10.1109/icati69399.2026.11604308
There is no free benchmark: An institutional view of legal AI benchmarking
Guha, N., Zhang, A. K., Tsang, C., Manning, C. D., Nyarko, J., & Ho, D. E. (2026) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2509757122 · Cited by 4
Self-citationMai, H., Li, W., Chen, J., Huang, B., Li, B., & Long, S. (2026) · International Conference on Computer and Multimedia Technology (ICCMT) · doi:10.1109/iccmt69755.2026.11608235
“𝑌𝑌′ This paper categorizes hallucinations in power writing into five types: entity hallucinations, numerical hallucinations, event hallucinations, procedural hallucinations, and citation hallucinations [4].”
Beyond silos: Anchoring trustworthy AI in cross-domain knowledge graphs for rural health resilience
Li, S., Guo, X., Juarez, F. G., Cao, J., Kulawiak, C., & Gong, J. (2026) · Smart Health · doi:10.1016/j.smhl.2026.100681
RATIO: A Graph-Aware Framework for Precedent-Grounded Legal RetrievalPreprint
Arop, J. (2026) · doi:10.14293/pr2199.004086.v1
Wu, J., Gan, L., Zhang, X., & Shu, Y. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7193839
The Division of Verification: Contractibility, Specialization, and the Limits of Delegation to AIPreprint
Zhang, H. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7086002
Applications of Generative AI for Access to Justice
Westermann, H., & Giacalone, M. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009779487.018
Making Copyright Law Computable: Rule-Based Text Extraction for WIPO Compliance Checking
Brüne, J. F., Corazza, M., & Palmirani, M. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-34304-8_3
Bias in AI: a framework for understanding the sources of bias in artificial intelligence
Robinig, W., & Wallner, J. P. (2026) · AI and Ethics · doi:10.1007/s43681-026-01283-0 · Cited by 3
“A generative AI system trained on such a corpus, without temporal weighting or domain-specific filtering, may reproduce the conceptual vocabulary of superseded legal regimes when asked to summarize, interpret, or generate legal content [ 5 , 7].”
The Future of Hybrid Intelligence in Administration and Digital Governance
Alibašić, H. (2026) · Public administration and information technology · doi:10.1007/978-3-032-28086-2_7
Análise de Falhas na Detecção de Alucinação em Textos Jurídico-Policiais em Português
Barcelar, R. R., & Ventura, T. M. (2026) · doi:10.5753/pesquisanuvem.2026.23945
Zhang, Y., Zhao, J., Ju, M., Ye, Q., Zhang, Y., Tang, K., Wang, Z., & Luo, H. (2026) · Artificial Intelligence and Law · doi:10.1007/s10506-026-09531-8
Nipa, S. N., & Taiba, J. F. (2026) · Dialogica · doi:10.5709/dialogica-02.s1.2026-06
Weber-Fares, R., Cochlin, F. J., Peddireddy, S. R., Howard-Williams, M., Sunshine, G., Ho, E., & Schmit, C. D. (2026) · The Journal of Law Medicine & Ethics · doi:10.1017/jme.2025.10224 · Cited by 1
Future-proofing agricultural research: FAIR principles for agriculture AI agents (FAIR4AG2)
Qian, C., Jung, Y., Hu, H., et al. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1877344
“Such failures are already documented beyond agriculture: leading models elaborated on fabricated clinical details inserted into decision-support prompts in 50–82% of cases (Omar et al., 2025), with comparably high hallucination rates reported for legal queries (Dahl et al., 2024).”
From Statute to Control Flow: Span-Grounded Deontic Trees for Defeasible Scope Parsing
Chen, J., Li, S., Wan, C., & Yuan, Z. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817491
Modernizing Heavily Regulated Industries with Autonomous Agentic AI: From LegalTech to InsurTech
Shashkov, V. (2025) · European Journal of Applied Science Engineering and Technology · doi:10.59324/ejaset.2025.3(2).31
Saltykov, A. (2026) · European journal of management, economics and business. · doi:10.59324/ejmeb.2026.3(1).25
Hạnh, V. T. (2026) · Multidisciplinary Reviews · doi:10.31893/multirev.2027032
“Building on this general risk profile, Dahl et al. (2024) document legal hallucinations — LLM outputs inconsistent with legal facts — in public-facing models and caution against the rapid, unsupervised integration of popular LLMs into legal tasks (Dahl et al., 2024).”
Saltykov, A. (2025) · Universal Library of business and economics. · doi:10.70315/uloap.ulbec.2025.0204017
Why Large Language Models Cannot Be Certified for Safety-Critical Systems
Sayakov, A. (2026) · American Impact Review · doi:10.66308/air.e2026066
Wang, J.-W., Tang, Y.-Z., Chen, X., Peng, H., & Zhang, Y.-L. (2026) · Journal of Hospitality and Tourism Management · doi:10.1016/j.jhtm.2026.101531
Agentic RAG for Legal Question Answering in Civil Law: Evidence From the Korean Bar Examination
Lim, E. Y., & Park, J. (2026) · IEEE Access · doi:10.1109/access.2026.3722717 · Cited by 1
“Dahl et al. [17] found that general-purpose LLMs produce hallucinated legal content on a substantial fraction of queries, and the problem persists even in purpose-built products: Magesh et al. [18] reported that commercial legal-research tools such as Lexis + AI and Westlaw AI-Assisted Research…”
Atencio, J. M. (2026) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.21970373
Human realignment: An empirical study of LLMs as legal decision-aids in moral dilemmas
Engel, C., Hermstrüwer, Y., & Kim, A. (2026) · Artificial Intelligence and Law · doi:10.1007/s10506-026-09517-6 · Cited by 6
The Migration of Bias in Law: From Human Cognition to Predictive AI System and Large Language Models
Zhong, H. (2026) · Springer eBooks · doi:10.1007/978-3-031-87993-7_185-1 · Cited by 1
ELARA: A Framework for Empirical Legal Research with AI AgentsPreprint
Choi, J. H. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7349078
Causal attributions for ChatGPT errors: what is the role of AI literacy?
Norcia, M., & Colí, E. (2026) · Journal of Psychology and AI · doi:10.1080/29974100.2026.2712302
Real-time triadic mediation: AI, epistemic symmetry, and preventive access to justice
Gazit, L. (2026) · AI and Ethics · doi:10.1007/s43681-026-01351-5
“Dahl et al. [4 ] provide the first systematic profile of legal hallucination in public-facing language models.”
Olthuis, E., Jansen, G., & Rietveld, R. (2026) · International Journal of the Legal Profession · doi:10.1080/09695958.2026.2710632
Beyond the autonomy illusion: hybrid intelligent systems and the architecture of AI accountability
Medsker, L., & Virmani, S. (2026) · AI and Ethics · doi:10.1007/s43681-026-01327-5
“Large language models generate fabricated legal citations with high confidence [ 6], produce fluent but factually incorrect clinical recommendations [ 8], and exhibit bias and opacity that resist correction because the relevant properties are distributed across billions of weights rather than…”
Green, D. (2026) · International Journal for the Semiotics of Law - Revue internationale de Sémiotique juridique · doi:10.1007/s11196-026-10572-0
Wang, X., & Lu, J. (2026) · Frontiers in Public Health · doi:10.3389/fpubh.2026.1937335
Duże modele językowe a prawo prywatne międzynarodowe
Świerczyński, M. (2025) · Problemy Prawa Prywatnego Międzynarodowego · doi:10.31261/pppm.2025.37.02
Lin, R., Zhang, L., Wang, H., Liao, X., Lu, Z., & Su, Y. (2026) · IEEE Access · doi:10.1109/access.2026.3731259
Why we believe chatbots: trust calibration as a design problem
Jaidka, K., & Cai, M. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1935527 · Cited by 4
“Beyond stylistic tropes, chatbots have been reported to produce fabricated but convincingly formatted evidence, including medical references (Alkaissi and McFarlane, 2023) and legal citations (Dahl et al., 2024), and hallucination is documented across model families (Ji et al., 2023).”
Ji, Q., Tang, J., Ye, L., & Wang, H. (2026) · Expert Systems with Applications · doi:10.1016/j.eswa.2026.134338
THEMIS: A Governance-first AI Help Desk for Court AdministrationPreprint
Panagiotis, P. E. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7410658
Rivera, B. K. G., Fernández, I. A., Cisneros, C. F. M., Bautista, J. U. Q., & Solano, C. A. G. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-37438-7_2
Aggarwal, S. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-32732-1_10
Lissack, M. R. (2026) · AI and Ethics · doi:10.1007/s43681-026-01372-0
“Recent empirical work on legal hallucinations [ 7, 17] provides evidence of the problem’s scope and suggests directions for intervention.”
Cavaleri, P. (2026) · JLIS it · doi:10.36253/jlis.it-749
“For example, Dahl et al. (2024), studying legal applications of large language models, documented frequent fabrication of legal citations.”
Paleari, M., Mangiapane, B., Vella, G. S., et al. (2026) · doi:10.1109/eeite69609.2026.11679791
“Specifically, the “black box” nature of neural models obscures transparency, making it difficult for human operators to establish appropriate levels of trust [3], while standard LLMs are documented to produce “legal fictions”, hallucinated, statutory details that lack any legal validity [6].”
The lawyer after AI: three horizons for the future of the legal profession
Farber, S. (2026) · International Journal of the Legal Profession · doi:10.1080/09695958.2026.2733071
Jayaratne, S. S., & Weerasinghe, R. (2026) · Moratuwa Engineering Research Conference · doi:10.1109/mercon71835.2026.11691544
“However, their application presents several challenges: Bezou-Vrakatseli et al. [11] highlights limitations in capturing formal argumentation schemes, while Dahl et al. [21] reports hallu-cination and factual inconsistency in generated outputs.”
Kamugisha, J. (2026) · AI and Ethics · doi:10.1007/s43681-026-01390-y
“Empirical research on public-facing large language models has shown that fluent legal responses can contain fabricated or incorrect authorities and that models may fail to recognise false assumptions embedded in a question [ 8 ].”
Evaluation of Prompt Engineering for LLM-Based Construction Contract Risk Analysis
Rasool, S., Chen, Y., Udeaja, C. E., & Fong, D. (2026) · Journal of Legal Affairs and Dispute Resolution in Engineering and Construction · doi:10.1061/jladah.ladr-1659
Zhang, Y., Li, Z., Kong, Z., & Xie, H. (2026) · Research Square · doi:10.21203/rs.3.rs-10925119/v1
程, 博. (2026) · Creative Education Studies · doi:10.12677/ces.2026.149743
Herrero, A. C., & Murillo, F. J. (2026) · Edutec Revista Electrónica de Tecnología Educativa · doi:10.21556/edutec.2026.97.4527
Prompt-Template Based Named Entity Recognition in Legal Domain: A Case Study with DeepSeek
Rai, L. (2026) · doi:10.1109/wsai70435.2026.11708695
“One prominent issue is hallucination, with studies reporting that large language models can generate spurious entities or misclassify legal terms at rates as high as 58%, undermining trust in extracted outputs [18].”
Xue, Y. (2026) · World Symposium on Artificial Intelligence (WSAI) · doi:10.1109/wsai70435.2026.11708627
“provides systematic evidence that public-facing LLMs hallucinate legal information and that hallucination rates vary across courts, jurisdictions, time periods, and cases [4].”
RAG-Based Intelligent Agent for Automated Generation of Organizational Information Security Policies
Lisov, A. S., & Breus, D. E. (2026) · International Russian Automation Conference (RusAutoCon) · doi:10.1109/rusautocon69876.2026.11707347
“Magesh et al. [10] and Dahl et al. [11] quantify factual error rates for Lexis+ AI (≈17”
Kulkarni, S., Aman Chandra, H., Saadhvi, S., Nithin, S., & S, S. M. J. (2026) · International Conference on Emerging Techniques in Computational Intelligence (ICETCI) · doi:10.1109/icetci71772.2026.11710573
“General-purpose LLMs trained on broad internet corpora may hallucinate legal provisions [4], confuse jurisdictions, overlook recent legislative amendments, and fail to provide traceable statutory citations, which are essential for legal reliability.”
Training datasets tailored for domain-specific large language models: challenges, methods and trends
Qiu, X., Ren, K., Cui, X., Liu, W., Gao, Z., Hu, J., & Liang, Q. (2026) · Journal of King Saud University - Computer and Information Sciences · doi:10.1007/s44443-026-01150-2
Closed-World Trustworthy Copilots for Legal and Regulatory Text
Ojha, M., & Mishra, P. S. (2026) · doi:10.1145/3843782.3844654
Çetinkaya, A. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.105225
Metadata
- Authors
- Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E.
- Affiliations
- Yale University (US)
- Stanford University (US)
- Corresponding author
- Dahl, Matthew
- Venue
- Journal of Legal Analysis
- Publisher
- Oxford University Press (OUP)
- Published
- 2024-01-01
- Volume
- 16
- Issue
- 1
- Pages
- 64-93
- ISSN
- 1946-5319 (ISSN-L)
- 2161-7201
- OpenAlex
- W4400047099
- Cited by
- 559
- References
- 158
- Citations by year
- 2026: 355 · 2025: 148 · 2024: 55
- FWCI
- 677.59
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- The Journal of Legal Analysis (publishedVersion, cc-by-nc)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Topics
- Social Sciences › Law › Artificial Intelligence in Law
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Law › Digitalization, Law, and Regulation
- Keywords
- large language models, ChatGPT, legal knowledge, legal reasoning, legal education, legal practice
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 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (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
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.