Imagine standing before a court and confidently relying on a legal rule that simply does not exist.

That is precisely what occurred before the High Court of England and Wales, where a lawyer submitted an AI-generated legal proposition that turned out to be fictitious. The incident, examined in Anthony Malcolm Cork & Anor v Mark Smith (2026), has become one of the clearest judicial warnings on the use of generative artificial intelligence in legal practice.

Although the case arose from an insolvency application, it raises a question that extends far beyond insolvency law:

When AI fabricates legal authorities or misstates the law, who bears responsibility?

The Court’s answer is unequivocal. While AI may assist lawyers in conducting research, drafting documents, and summarising complex materials, responsibility for the accuracy of legal submissions remains with the human user.

From AI Hallucinations to Process-Based Accountability

The facts of Cork v Smith illustrate the risks associated with the use of generative AI in legal practice. A junior solicitor relied on an AI-generated version of Insolvency Rule 12.37(5), which appeared to confer a power on the court but in fact did not exist. The fabricated provision was then included in correspondence sent to the court without adequate verification.

Judge Mullen correctly focused on the conduct of the lawyers rather than the AI system itself. The judgment noted that the AI had warned users to verify its outputs against authoritative legal sources, yet those warnings were ignored.

This approach is consistent with decisions such as Mata v Avianca Inc (2023) and R (Ayinde) v London Borough of Haringey (2025), where courts emphasised that lawyers remain responsible for checking legal authorities.

The judgment was therefore correct in principle: professional accountability must remain with human lawyers, as public confidence in the legal system depends on identifiable and responsible decision-makers.

The Comparative Law Gap: Existing Professional Rules Are Not Designed for Generative AI

Although Cork v Smith reached the correct outcome, the judgment also highlights a deeper problem.

The Court relied on traditional professional duties contained within the Solicitors Regulation Authority Code of Conduct, including duties of competence, supervision, and honesty. These obligations undoubtedly apply to AI-assisted legal work. However, they were not designed with generative AI in mind.

The comparative position demonstrates this gap clearly.

In England and Wales, there is currently no explicit duty requiring solicitors to understand the limitations of AI systems before using them. Instead, regulators rely upon general obligations to provide competent legal services. Similarly, Ayinde treated AI misuse as a breach of existing professional duties rather than recognising any distinct obligations associated with AI use.

The United States has moved slightly further. Comment 8 to Rule 1.1 of the American Bar Association’s Model Rules of Professional Conduct requires lawyers to maintain competence regarding “the benefits and risks associated with relevant technology.” This has become known as the duty of technological competence.

However, even this doctrine emerged before the rise of generative AI and does not specifically address hallucinations, fabricated authorities, or prompt engineering. Cases such as Mata v Avianca Inc therefore continue to rely on traditional competence principles rather than AI-specific standards.

The European Union’s AI Act similarly focuses on deployers, providers, and risk management systems rather than legal professionals. While it imposes extensive obligations concerning high-risk AI systems, it offers little guidance regarding the professional responsibilities of lawyers who use generative AI in litigation.

The result is a paradox.

Courts increasingly sanction lawyers for AI misuse, yet few jurisdictions have clearly defined comprehensive standards for competent AI use in legal practice. Existing frameworks identify failures after they occur but provide limited guidance regarding how lawyers should avoid those failures in the first place.

This regulatory gap is particularly concerning because generative AI differs fundamentally from traditional legal research tools.

Databases such as Westlaw and Lexis retrieve existing authorities. By contrast, large language models generate text probabilistically and may fabricate sources altogether. As Bender and her co-authors have argued, such systems are designed to produce plausible language rather than verified truth.

Treating them as equivalent to traditional research tools therefore obscures the distinct risks they create.

Towards a Duty of AI Competence

Cork v Smith should therefore be understood not merely as a case about professional negligence but as evidence that existing legal ethics frameworks require reform.

The next stage of AI governance should involve recognising a distinct duty of AI competence. Such a duty would not require lawyers to become computer scientists. Rather, it would establish minimum standards for the responsible use of generative AI in legal practice.

At a minimum, the duty should include five obligations:

1. Lawyers should understand the limitations of generative AI, including hallucinations, bias, and unreliable legal citations.
2. AI-generated authorities, quotations, and statutory references should always be verified against authoritative primary sources before they are relied upon in legal advice or presented to a court.
3. Law firms should adopt formal AI governance policies supported by regular training, identifying approved AI tools, appropriate use cases, supervisory responsibilities, and mandatory human review.
4. Lawyers should maintain an internal record of significant AI use during the preparation of litigation documents, creating an audit trail for compliance and accountability.
5. Courts and professional regulators should consider requiring disclosure where generative AI has materially influenced legal submissions or evidentiary analysis.

These obligations would move professional regulation beyond reactive sanctions and towards proactive governance.

Rather than merely punishing lawyers after AI failures occur, regulators would provide clear standards capable of guiding responsible behaviour.

Conclusion

Cork v Smith is not merely a case about AI-generated hallucinations or professional negligence.

Its lasting significance lies in exposing a structural mismatch between twentieth-century legal ethics and twenty-first-century legal technology.

The Court rightly reaffirmed that responsibility for legal submissions cannot be delegated to an algorithm. However, assigning responsibility after an error occurs is only part of the regulatory response. Equally important is defining what responsible AI use requires before mistakes arise.

That task now falls to professional regulators.

Rather than continuing to rely exclusively on broad duties of competence and supervision, bar associations and law societies should articulate AI-specific standards governing verification, training, firm-level governance, record-keeping, and, where appropriate, disclosure of material AI use.

Such standards would not diminish lawyers’ professional autonomy; they would provide the clarity needed to exercise that autonomy responsibly in an era of generative AI.

As artificial intelligence becomes embedded in everyday legal practice, the central question is no longer whether lawyers may use AI, but what obligations accompany its use.

Cork v Smith marks the beginning of that conversation. Whether it leads to a coherent framework of AI competence will depend not on the courts, but on the willingness of regulators to modernise legal ethics for the age of generative AI.

References

1. Anthony Malcolm Cork & Anor v Mark Smith (2026). EWHC 1199 (Ch) https://www.bailii.org/ew/cases/EWHC/Ch/2026/1199.html
2. Mata v Avianca Inc (2023). 678 F Supp 3d 443 (SDNY) https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/
3. R (Ayinde) v London Borough of Haringey (2025). EWHC 1383 (Admin) https://www.judiciary.uk/judgments/ayinde-v-london-borough-of-haringey-and-al-haroun-v-qatar-national-bank/
4. Frank Pasquale (2015). The Black Box Society: The Secret Algorithms That Control Money and Information https://www.jstor.org/stable/j.ctt13x0hch
5. American Bar Association (2020). Model Rules of Professional Conduct, Rule 1.1, Comment 8 https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/
6. European Parliament and Council (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act) https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
7. Emily M Bender, Timnit Gebru, Angelina McMillan-Major & Margaret Mitchell (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? https://s10251.pcdn.co/pdf/2021-bender-parrots.pdf
8. Deborah Lupton / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/