Citing Ghosts: AI Hallucinations And The Future Of Legal Responsibility

We are living in an era in which digitalization is accelerating and Artificial Intelligence ("AI") technologies are reshaping both societal and professional practices. In the age of generative AI, where new developments emerge almost daily, the legal industry cannot remain unaffected by this transformation. As of 2025, AI is no longer a distant prospect for future debate; it has become an auxiliary tool and a strategic component in the practical work of legal professionals.

According to a 2025 study by Thomson Reuters¹, 80% of legal professionals believe that AI will have a significant or transformative impact on their profession within the next five years. The findings of the study demonstrate that AI delivers notable gains in speed, cost efficiency, and productivity, particularly in areas such as legal research, document review, e-discovery, and data management. Yet, alongside these advantages, the use of AI raises pressing concerns regarding accuracy, transparency, trust, and ethical responsibility. In this regard, the black box nature of AI and the increasingly debated phenomenon of AI hallucinations compel a re-examination of the boundaries of professional responsibility in legal practice.

This article will first briefly address the question of what AI hallucination is and how it gives rise to contemporary legal concerns, and then conduct a concise comparative overview of some of the commonly cited court cases, ultimately proposing a limited guideline for Turkish law to adopt to ensure the safe and efficient use of AI.

When Machines Imagine: Understanding AI Hallucinations

Artificial intelligence systems, particularly large language models ("LLMs"), are often characterized as "black boxes." Their internal decision-making processes remain opaque to the user, and it is often unclear which data, training inputs, or weight combinations directly influence the output.

AI hallucination emerges as a phenomenon closely linked to this black-box structure. It refers to situations where AI systems produce outputs that are unfounded, misleading, or entirely fabricated; often presenting fictitious sources, case law, or legal norms in a convincing manner, thereby creating a false impression of reliability for the user.

A recent study has demonstrated that even the most advanced AI tools developed for legal research are not entirely "hallucination-free"; rather, they generate inaccurate or unfounded references at rates ranging from 17% to 33%.² The manifestation of this phenomenon in legal practice is most evident in the use of fabricated case law in pleadings, errors within data submitted to courts, and reliance on flawed authorities in decision-making processes. Accordingly, AI hallucination is not merely a technical issue but a critical concern directly implicating the accuracy of judicial proceedings and the integrity of professional ethics. The AI Hallucination Cases Database - created and maintained by Damien Charlotin³, Senior Research Fellow at HEC Paris and lecturer at Sciences Po with a research focus on AI and the law- systematically tracks this phenomenon, recording cases in which fabricated case law and erroneous arguments generated by AI tools have come under judicial scrutiny. As of September 2025, 383 cases have been identified, and the database continues to expand as new examples emerge.⁴

Comparative Overview: Judicial Perspectives on AI Hallucinations

At the international level, it is evident that judicial bodies have not confined their responses to AI-related "hallucination" incidents to retrospective disciplinary sanctions, but have also developed preventive and procedural measures. In the United States, for instance, several federal courts have issued standing orders and local rules requiring attorneys to explicitly disclose any use of generative AI and to personally verify the accuracy of all legal citations. In this regard, Judge Michael Baylson's (E.D. Pa., June 2023) ⁵ disclosure requirement and the parallel orders issued by Judges Gabriel Fuentes and Jeffrey Cole (N.D. Ill., 2023/24)⁶ stand out as innovative and proactive procedural measures. Such directives adapt the "reasonable inquiry and verification obligation" embodied in Rule 11 of the Federal Rules of Civil Procedure to the realities of the AI era, thereby aiming to prevent fabricated case law from entering the record and to safeguard the integrity of judicial proceedings.

The number of judicial decisions worldwide that address AI hallucination is steadily increasing, often accompanied by the introduction of various sanctions. In this context, the approaches adopted across different legal systems warrant a comparative look. Now, let's take a look at some recent key cases.

• Mata v. Avianca (S.D.N.Y, 2023 - USA) ⁷
  One of the earliest landmark cases in which the use of artificial intelligence was directly contested in legal practice is Mata v. Avianca, which attracted wide attention in 2023. On February 2, 2022, Roberto Mata initiated proceedings before the New York State Supreme Court. In responding to the defendant airline's statute of limitations defense, the plaintiff's counsel submitted a brief that incorporated responses generated by ChatGPT without independently verifying their accuracy. These submissions included case law that did not, in fact, exist, as well as misquoted authorities. In its decision dated June 22, 2023, the court sanctioned plaintiff's counsel for failing to exercise the requisite professional diligence. Pursuant to Rule 11 of the Federal Rules of Civil Procedure⁸, the court imposed a USD 5,000 monetary sanction on the attorneys and ordered them to send letters to the judges and clients, explaining the circumstances of the misrepresentations.
  The court emphasized that the mere use of a reliable AI tool is not inherently problematic; however, counsel bears an independent duty to ensure the accuracy of any submissions made to the court. Moreover, the sanction was directed not only at the attorney who prepared and signed the brief but also at the law firm where the counsel worked, holding them jointly and severally liable for the monetary penalty, thereby underscoring the institutional responsibility of law firms.
• Lacey v. State Farm (C.D. Cal., 2025 - USA)⁹
  In this insurance dispute, the plaintiff's counsel relied on AI-generated fabricated court decisions, and the same errors persisted even in revised pleadings despite the inaccuracies being identified. The court characterized this conduct as tantamount to bad faith and imposed monetary sanctions totaling USD 31,100 on the law firms, along with an order to reimburse the opposing party's expenses. In addition, the plaintiffs' supplementary requests were struck from the record. This decision is noteworthy not only for the monetary sanctions but also for its procedural impact, as the denial of ancillary requests directly influenced the course of the litigation.
• Ayinde v. London Borough of Haringey; Al-Haroun v. Qatar National Bank (High Court, 6 June 2025 - UK)¹⁰
  The High Court of England and Wales examined the inclusion of AI-generated fabricated court decisions in two separate cases, characterizing such conduct not only as a professional breach but also as a threat to the integrity of the judiciary. In the Haringey case, the court imposed a wasted costs order of £2,000 and, in both cases, referred the responsible counsel to the Solicitors Regulation Authority (SRA)¹¹ and the Bar Standards Board (BSB).¹² The decision underscored that the use of AI in legal practice is acceptable only when accompanied by independent verification processes and that serious violations may fall within the ambit of "perverting the course of justice."
• Garner v. Kadince (Utah Court of Appeals, 2025 – USA)¹³
  The Utah Court of Appeals imposed disciplinary and monetary sanctions on attorneys Douglas Durbano and Richard Bednar for submitting a brief, prepared by a law student intern that contained fabricated case law generated by ChatGPT without independent verification. Referring to Mata v. Avianca, the court emphasized that the use of AI tools does not relieve counsel of their duty of verification, noting that this obligation rests with the signing attorney and cannot be delegated. As part of the sanctions, Bednar was ordered to reimburse opposing counsel for the costs incurred in responding to the brief, to refund the fees collected from their client, and to donate USD 1,000 to "And Justice for All¹⁴ " within 14 days. This decision demonstrates that AI-related errors may result not only in monetary penalties but also in financial accountability toward the client and obligations of social contribution.
• Tribunale di Torino, Sezione Lavoro - No. 1018/2025 (2025 - Italy)¹⁵
  The Turin Labour Court examined a pleading drafted with the assistance of AI, which consisted largely of irrelevant and fabricated legal references unrelated to the dispute over social security payment orders. The court held the claims manifestly unfounded and emphasized that the use of AI does not excuse, but rather highlights, a breach of diligence. As a result, the counsel's claims were dismissed, and the counsel was ordered to reimburse the opposing parties' litigation costs. In addition, the court imposed further sanctions on the counsel, requiring payment of €500 to each opposing party and €500 to a public fund administered by the Ministry of Justice, explicitly on the ground that the unverified use of AI-generated content rendered the pleading frivolous and abusive.

Turkish Legal Perspective: the Multi-Layered Responsibility of Attorneys

In Turkish law, when the strict procedural rules governing pleadings and submission of evidence are considered together with the attorney's duties of diligence, candor, and loyalty -giving rise to professional, contractual, and criminal liability- the risks posed by the use of artificial intelligence must be addressed within a multi-layered regime of responsibility.

Legal Framework for Submission of Evidence and Duty of Candor

The Turkish Code of Civil Procedure numbered 6100 ("TCCP")¹⁶ provides a detailed framework governing the content of pleadings and submission of evidence. Accordingly, a statement of claim must expressly set out the facts underlying the allegation, the evidence by which such facts will be proven, and the legal grounds relied upon. In this context, the parties are required to substantiate their claims in a conducive manner and to correlate the evidence with the alleged facts.¹⁷ The use of AI-fabricated evidence will primarily result in such evidence being struck out of the case, thereby rendering the argument brought forward groundless.

TCCP also imposes on parties a duty to act in accordance with the principle of honesty and to ensure the truthfulness of their submissions.¹⁸ A breach of this obligation may also result in the relevant party bearing all or a part of the litigation costs, excluding the judgment fee, regardless of whether the case is ultimately decided in their favor.¹⁹

Liability Regime for Attorneys

Apart from the procedural obligations imposed directly on the parties, the responsibility of attorneys also requires separate consideration. The role of the attorney encompasses professional duties grounded in diligence, candor, and loyalty. In Turkish law, the scope of an attorney's liability may be examined under three categories: (i) professional liability under professional rules and disciplinary standards; (ii) liability towards the client, stemming from the attorney–client contractual relationship; and (iii) criminal liability in cases where the conduct of the attorney constitutes a criminal offense.

Professional liability of attorneys is framed by the Attorneyship Law ("AL")²⁰ and the Union of Turkish Bar Associations' Code of Professional Conduct ("UTBA Code of Conduct"). AL requires attorneys to perform their duties with diligence, integrity, and honor; to act in a manner befitting the trust inherent in the professional title, and to comply with professional rules²¹. This principle is linked not only to legal expertise but also to the public trust dimension of the profession. UTBA Code of Conduct also requires professional work to be conducted faithfully and in a manner that sustains public confidence; it prohibits conduct that would damage the profession's reputation. The framework established by the AL and the UTBA Code of Conduct is directly applicable to errors arising from AI-generated "hallucinations." In particular, the inclusion of fabricated court decisions in pleadings could constitute a breach of the duty of diligence and the obligation to preserve the integrity of the profession. Such conduct would originally trigger disciplinary liability ranging from official warnings and monetary penalties to disbarment; however, in its decision of 6 March 2025, the Constitutional Court annulled the provisions regulating such sanctions, thereby creating the need for a future re-regulation of the disciplinary framework.²²

Attorneys are also liable towards their clients under the attorneyship provisions of the Turkish Code of Obligations numbered 6098 ("TCO")²³, which require them to act with loyalty and diligence while safeguarding the client's interests. In this context, submitting fabricated or misleading documents - including AI-generated fictitious court decisions- may be considered a serious breach of the duties of loyalty and diligence, potentially giving rise to attorneys' liability in damages.

Finally, attorneys' criminal liability under the Turkish Penal Code numbered 5237 ("TPC")²⁴ must be addressed in light of the fact that the legal profession is defined as a public service under Turkish law²⁵, which in turn subjects attorneys to provisions applicable to public officials in the event of criminal offenses. Within this framework, an attorney's neglect of duty or failure to properly discharge professional obligations may, under certain circumstances, be associated with the offense of "abuse of office"²⁶ as regulated under the TPC. In particular, where AI-generated erroneous or fictitious court decisions are submitted to the court without verification, thereby causing harm to individuals, damage to the public, or granting undue advantage to third parties, such conduct may potentially fall within the scope of this crime.

Looking Ahead: Reflections and Prospective Pathways for Turkish Law

Examples from foreign legal systems demonstrate that errors stemming from AI hallucinations, such as fabricated court decisions and misattributed citations, are not merely individual attorney lapses but a structural issue that threatens the integrity of judicial proceedings. Although no case of this nature has yet been adjudicated in Türkiye, the sanctions and preventive measures emerging in comparative law suggest that the development of internal regulatory mechanisms to address similar risks is necessary. In this regard, the principal aspects for Turkish law can be summarized as follows:

• Professional Diligence and Disciplinary Dimension: The duty of diligence set out in AL may equally come into play in cases of violations arising from the use of artificial intelligence. In light of the Constitutional Court's annulment decision and the resulting regulatory gap, it is important that disciplinary sanctions be reconsidered so as to encompass such violations and clarified within the framework of professional conduct rules.
• Procedural Framework: Within the TCCP and the Turkish Code of Criminal Procedure, discussions could be initiated on procedural regulations requiring disclosure of the use of AI-based tools in pleadings and mandating that attorneys personally verify the accuracy of citations, similar to the standing order examples seen in the United States.
• Institutional Responsibility and Education: It would be valuable for law firms to establish internal compliance and accountability mechanisms, for bar associations and professional bodies to promote awareness through continuing legal education on AI, and, in the longer term, for law faculties to incorporate courses addressing the relationship between AI and law into their curricula. These measures could play an important role in raising awareness and strengthening the professional responsibility of legal practitioners.

Finally, it may be anticipated that Türkiye, in due course, will also move toward comprehensive legislation similar to the European Union's AI Act²⁷, addressing the legal liability dimension of artificial intelligence. Such an initiative would not only influence the practice of law but could also signal the emergence of a preventive and holistic approach to safeguarding the integrity of judicial processes more broadly.

Footnotes

1. Thomson Reuters, How AI is Transforming the Legal Profession (21 March 2025), available at: https://legal.thomsonreuters.com/blog/how-ai-is-transforming-the-legal-profession/

2. Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D Manning ve Daniel E Ho, Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (arXiv preprint, 2024) https://arxiv.org/pdf/2405.20362

3. Damien Charlotin, Curriculum Vitae, available at: https://www.damiencharlotin.com/cv/

4. Damien Charlotin, AI Hallucination Cases Database https://www.damiencharlotin.com/hallucinations/

5. Standing Order Re Artificial Intelligence, Eastern District of Pennsylvania (U.S. District Court), 6 June 2023, https://www.paed.uscourts.gov/sites/paed/files/documents/procedures/Standing%20Order%20Re%20Artificial%20Intelligence%206.6.pdf

6. Standing Order for Civil Cases Before Judge Gabriel A. Fuentes (rev'd May 31, 2023), U.S. District Court for the Northern District of Illinois, https://www.ilnd.uscourts.gov/_assets/_documents/_forms/_judges/Fuentes/Standing%20Order%20For%20Civil%20Cases%20Before%20Judge%20Fuentes%20rev%27d%205-31-23%20%28002%29.pdf

7. Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), 2023 WL 4114965 (S.D.N.Y. June 22, 2023), https://cases.justia.com/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/0.pdf?ts=1687525481

8. Federal Rules of Civil Procedure. Rule 11: Signing pleadings, motions, and other papers; representations to the court; sanctions. Legal Information Institute. https://www.law.cornell.edu/rules/frcp/rule_11

9. Lacey v. State Farm, C.D. Cal., 2025, https://s3.documentcloud.org/documents/25940991/lacey-v-state-farm.pdf

10. Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank, High Court of Justice (England and Wales, 6 June 2025), https://www.judiciary.uk/wp-content/uploads/2025/06/Ayinde-v-London-Borough-of-Haringey-and-Al-Haroun-v-Qatar-National-Bank.pdf

11. Solicitors Regulation Authority (SRA), the regulatory body for solicitors in England and Wales, https://www.sra.org.uk/

12. Bar Standards Board (BSB), the regulator of barristers and specialised legal services businesses in England and Wales, https://www.barstandardsboard.org.uk/

13. Garner v. Kadince, Utah Court of Appeals, No. 20250188-CA (Utah Ct. App. published, 2025), https://cases.justia.com/utah/court-of-appeals-published/2025-20250188-ca.pdf?ts=1748025697.

14. And Justice for All, a Utah-based nonprofit providing free or low-cost legal services to vulnerable groups -including low-income individuals, people with disabilities, the elderly, and victims of domestic violence- through partnerships with the Disability Law Center, Utah Legal Services, and the Legal Aid Society of Salt Lake, serving approximately 25,000 people annually, available at https://andjusticeforall.org

15. Sentenza Tribunale di Torino, Sezione Lavoro, Sent. n. 1018/2025 (16 September 2025), https://websitedc.s3.amazonaws.com/documents/AI_Allucination_-_Turin_Court_n._1018-2025.pdf

16. Published in the Official Gazetted dated 4 February 2011 and numbered 27836.

17. Art. 119 and 194 TCCP.

18. Art. 29 TCCP.

19. Art. 327 TCCP

20. Published in the Official Gazette dated 7 April 1969 and numbered 13168.

21. Art. 34 LA

22. In its decision dated 6 March 2025 (E.2025/50, K.2025/47), the Constitutional Court annulled Articles 134 and 135 of the Attorneys' Act on the grounds that they violated the principles of legal certainty and proportionality. The Court emphasized, in particular, that the lack of foreseeability as to which acts corresponded to which disciplinary sanctions, coupled with the broad and unchecked discretion vested in bar association bodies, contravened the rule of law principle (Constitution, Art. 2). The decision was published in the Official Gazette on 22 May 2025, No. 32907. See Official Gazette, "Constitutional Court Decision," 22.05.2025, No. 32907 https://www.resmigazete.gov.tr/22-05-2025

23. Published in the Official Gazetted dated 4 February 2011 and numbered 27836.

24. Published in the Official Gazetted dated 12 October 2004 and numbered 25611.

25. Art 1 AL

26. Art 257 TPC.

27. European Parliament and Council, Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending certain Union legislative acts (Artificial Intelligence Act), OJ L, 2024/1689, 12.7.2024, p. 1–132. Available at: https://eur-lex.europa.eu/eli/reg/2024/1689/oj

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.