The DOI matches its Crossref record; 158 of the 158 references it cites were found in a registry or an index; 29 of the 52 sources its claims cite were checked for support.
Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models
Open accessAbstract
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.
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 2023Experimental — 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 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)
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) 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 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 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 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 source
- Partly supportedcites Savelka et al. 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: 8
- Ambrogi, 2023, As Allen & Overy Deploys GPT-based Legal App Harvey Firmwide, Founders Say Other Firms Will Soon Follow, LawSites
- Chien, 2024, How Generative AI Can Help Address the Access to Justice Gap Through the Courts, Loyola Los Angeles Law Rev.
- Draper, 2023, The Potential for Jurisdictional Challenges to AI or LLM Training Datasets
- Dworkin, 1986, Law’s Empire
- Mayson, 2019, Bias In, Bias Out, Yale Law J., 128, 2122
- Romoser, 2023, No, Ruth Bader Ginsburg Did Not Dissent in Obergefell — and Other Things ChatGPT Gets Wrong about the Supreme Court, SCOTUSblog
- Simshaw, 2022, Access to A.I. Justice: Avoiding an Inequitable Two-Tiered System of Legal Services, Yale J. Law Technol., 24, 150
- Weiser, 2023, Here’s What Happens When Your Lawyer Uses ChatGPT, New York Times
Cited by
AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges
Sapkota, R., Roumeliotis, K. I., & Karkee, M. (2025) · arXiv · doi:10.48550/arxiv.2505.10468 · Cited by 295
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
“In high-stake domains such as legal consultation or scientific research, these hallucinations can lead to severe misjudgments and erode user trust [190,191].”
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
Hallucination‐Free? Assessing the Reliability of Leading AI Legal Research Tools
Magesh, V., Surani, F., Dahl, M., Suzgun, M., Manning, C. D., & Ho, D. E. (2025) · Journal of Empirical Legal Studies · doi:10.1111/jels.12413 · Cited by 182
Self-citationGenerative AI in EU law: Liability, privacy, intellectual property, and cybersecurity
Novelli, C., Casolari, F., Hacker, P., Spedicato, G., & Floridi, L. (2024) · Computer Law & Security Review · doi:10.1016/j.clsr.2024.106066 · Cited by 124
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models. https://app.cochat.ai/papers/10.1093/jla/laae003
- The paper
- Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E. (2024). Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models. Journal of Legal Analysis. https://doi.org/10.1093/jla/laae003
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.