A junior fee earner uses artificial intelligence to research a point of law. The results look convincing, complete with case names, citations and apparently relevant legal principles. The research is incorporated into court documents and approved by a supervising solicitor.

There is just one problem. Some of the authorities do not exist.

For solicitors, this is no longer a hypothetical scenario. In UK and R (Munir) v Secretary of State for the Home Department [2026] UKUT 81 (IAC), promulgated on 17 November 2025 and published on 19 February 2026, the Upper Tribunal dealt with fabricated and inaccurate authorities in two cases heard together. It said that lawyers who put false authorities before a court or tribunal are likely to be referred to their regulator. It also said that a supervisor who fails to ensure a junior colleague’s work is free from false authorities is likely to be more culpable than a lawyer who fails to check their own work.

The decision arose in immigration proceedings, but the principles apply to every firm that relies on delegated research, junior fee earners or AI-assisted drafting. The SRA has since cited the case in its warning notice on the misuse of AI, published on 17 August 2026.

Who is more culpable when fake case law reaches a tribunal? The Tribunal said a supervisor who fails to ensure a junior’s work contains no false cases or citations is likely to be more culpable than a lawyer who fails to ensure their own work is free from them. Supervisors who fail in this way, or who fail to check a junior’s drafting, are likely to be referred to the SRA.

What did the Upper Tribunal decide?

The Tribunal was exercising its Hamid jurisdiction, which allows it to examine whether legal representatives have met their professional obligations to the court. The jurisdiction takes its name from R (Hamid) v Secretary of State for the Home Department [2012] EWHC 3070 (Admin).

The first case concerned grounds of appeal citing a Court of Appeal decision, Horleston v SSHD, that does not exist. The citation given belongs to an unrelated equal pay case. The adviser said he had not used AI to draft the grounds but could not explain how the false case got in. His best guess was that he had inadvertently used the “AI Mode” of a Google search. The Tribunal found no reported case of that name. It showed that Google’s AI could produce invented panels of Court of Appeal judges for it, with a different bench each time the question was reworded. It made no referral because the adviser had already reported himself to his regulators.

The second case concerned grounds for judicial review. Two Upper Tribunal decisions had been given High Court citations, a third was described as a Court of Appeal decision, and a fourth was cited under a citation that does not exist. The solicitor named in the claim form’s statement of truth was also the firm’s COLP. He said a part-time junior member of staff had drafted the grounds under his supervision, relying on an outdated precedent, practitioner blogs and personal notes without checking them against official sources. The Tribunal accepted that he had fifteen years of unblemished practice and that he regretted what had happened. It concluded, however, that the false citations in the grounds, and his failure to supervise the work in this and an unknown number of other cases, required a referral to the SRA.

The Tribunal’s general conclusions begin with two points: lawyers must ensure the arguments they present are factually and legally accurate, and citing false cases wastes the Tribunal’s time.

At paragraph 58, it stated:

“A solicitor or other legal professional who delegates their work to another fee-earner remains responsible for the supervision of their work and for ensuring its accuracy.”

Paragraph 58 adds that supervisors must ensure the fee earners they supervise are aware of the dangers of using non-specialist AI for legal research and drafting. Failing to do so, or failing to check their drafting, is likely to lead to referral. In paragraph 38 the Tribunal explained why supervisors face the greater criticism. A lawyer who fails to check their own work fails the Tribunal, the public and the client. A supervisor does all of that and also fails to help develop more junior lawyers.

Why this is a supervision case, not only an AI case

The Tribunal said it would be easy to see the second case as one about the naive use of generative AI, but that it is principally about supervision and the duty not to mislead the tribunal. It added that it matters not how citation errors come about, whether through a trainee or through ChatGPT. The qualified professional with conduct of the matter is expected to ensure that documents are checked and only accurate documents are filed.

The hearing brought out several practical points:

• AI is not confined to chatbots. The COLP told the Tribunal that his staff had no means of using AI. The Tribunal pointed out that anyone with access to Google has access to AI, and that he had taken no steps to warn staff about the pitfalls.
• A named supervisor is not evidence of supervision. The grounds were unsigned and bore only the firm’s name. At the permission stage, a judge ordered the COLP to identify the author. At the hearing, the COLP could not say who had completed the claim form, and the spreadsheet he produced to show who had worked on the file had no entry for the case. The Tribunal listed record-keeping among several concerns about the firm’s procedures.
• A family illness did not explain the failure in that case. The Tribunal did not accept it as an adequate explanation for the lack of checks. Among other things, the evidence did not show that the difficulty existed when the proceedings were being prepared.

The Tribunal built on R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin). There the Divisional Court warned that freely available generative AI tools cannot conduct reliable legal research. They can produce plausible responses that cite sources that do not exist, or attribute passages to genuine judgments that do not contain them. The Tribunal also recorded a considerable increase in fictitious citations in the second half of 2025.

The SRA’s effective supervision guidance, updated on 12 June 2026 after the Court of Appeal’s Mazur decision, says that putting supervision arrangements in place is necessary but not sufficient. Firms must take proactive steps to make sure supervision operates effectively. The guidance also addresses AI directly. It says outputs produced with AI assistance must be subject to appropriate human review, scrutiny and professional judgement, and that an authorised individual must retain ultimate responsibility for services delivered with AI assistance. For the wider context on Mazur, see New SRA supervision guidance post-Mazur: what’s changed and why culture is the real test.

A supervising solicitor cannot demonstrate adequate oversight simply by confirming that they have read a draft. Where authorities are relied upon, the process must be capable of establishing that they exist, are correctly cited and support the propositions advanced. That does not mean supervisors must repeat every step of a junior colleague’s research. It does mean the firm needs a reliable verification process, with clear accountability.

Three checks that every court submission should satisfy

The Tribunal has amended its judicial review claim form. The judgment records the change, which responds to the rise in fictitious citations, and says other forms and directions will be amended in the same way. Both cases before the Tribunal concerned pleadings settled before the change.

Legal representatives must now confirm, through a statement of truth, that any authority cited in the form or in documents appended to it:

1. Exists: The judgment or legal authority is real.
2. Can be located: The citation provided enables the authority to be found.
3. Supports the proposition: The authority establishes or supports the legal argument for which it is cited.

These are three distinct checks. Finding a case with the correct name does not establish that the citation is accurate. Confirming that a real judgment exists does not mean it says what the document claims.

The Tribunal warned that a representative who signs the statement of truth where false authorities have been cited should ordinarily expect referral to their regulator. The requirement applies to the Upper Tribunal’s judicial review process, but it works as a practical minimum standard for any firm reviewing how authorities are checked before filing.

What are the SRA obligations for solicitors using AI?

AI does not create a separate category of professional responsibility. Solicitors and firms remain subject to the SRA Standards and Regulations whether work is done by hand, delegated or assisted by technology. The relevant duties include:

• Duty to the court: Not misleading the court or tribunal. The Tribunal stressed that a lawyer’s primary duty is to the court and to the cause of truth and justice.
• Competence: Making sure that those you manage are competent to carry out their role (paragraph 3.6 of the Code for Solicitors).
• Supervision: Remaining accountable for work carried out through others and effectively supervising work done for clients (paragraph 3.5 of the Code for Solicitors). The Code for Firms (paragraph 4.4) requires an effective system for supervising client matters.
• Honesty, integrity and public trust: Maintaining the standards that underpin confidence in the profession.
• Confidentiality: Protecting client information.
• Governance: Maintaining effective governance structures, systems and controls (paragraph 2.1 of the Code for Firms). The COLP must take all reasonable steps to ensure compliance, including with supervision requirements (paragraph 9.1).

The SRA’s warning notice on the misuse of AI, published on 17 August 2026, identifies false citations and client confidentiality as its two areas of concern. It cites Ayinde and this decision as evidence of both. The SRA reportedly received 42 reports of potential AI misuse between July 2025 and July 2026, with investigations continuing. For more on the notice, see SRA warning notice puts AI governance and client confidentiality into the spotlight.

For COLPs, the question is not only whether individual fee earners understand AI’s limitations. It is whether the firm has arrangements to stop those limitations becoming breaches of professional obligations.

Five practical steps for COLPs and managing partners

1. Review how fee earners conduct legal research

Establish which AI tools are being used across the firm and for what. Are trainees, paralegals and other fee earners using general-purpose tools or search-engine AI features to find case law? Are they relying on AI summaries without reading the original judgments? Do supervisors know when AI has contributed to research?

A firm cannot manage risks it has not identified. Consider requiring staff to disclose AI assistance in legal research and to distinguish verified primary sources from unverified AI output. AI features now sit inside ordinary search engines, so a ban on named chatbots will not cover the risk.

2. Introduce a documented authority verification process

Where research supports advice, pleadings, witness statements or submissions, verification should be part of the normal workflow. Check authorities against official judgments or authoritative legal databases. Establish that the case exists, the citation is correct, the relevant passage has been accurately represented and the authority supports the argument. For higher-risk work, consider requiring a supervising solicitor to approve the authorities before documents are finalised. Verification must be demonstrable, not assumed.

3. Strengthen supervision and record-keeping for delegated work

Consider whether:

• Junior fee earners receive clear instructions about the research they undertake.
• Supervisors know which tasks have been delegated and whether AI has been used.
• Approval processes distinguish between checking presentation and verifying legal substance.
• The firm can show who drafted each document, who reviewed it and which authorities were checked, from records that are kept reliably.
• Work on unfamiliar or complex issues receives proportionately greater scrutiny.

The Tribunal’s questioning of the COLP in Munir shows how unhelpful it is when a firm cannot say who worked on a file. These measures should sit within the firm’s wider supervision arrangements, not operate as a separate, informal AI check.

4. Provide practical training and warn staff directly

Paragraph 58 makes it a supervisor’s responsibility to ensure fee earners understand the dangers of non-specialist AI. A written policy alone will not discharge that. Training should use examples of fabricated citations, misleading summaries of real judgments and incorrect quotations, and should cover the AI features built into search engines. Supervisors need training too, on the tools in use and on what an adequate review of AI-assisted work involves. Policies also need to keep pace with the technology, as set out in AI policies in law firms need constant attention.

5. Build AI risks into compliance monitoring and file reviews

Consider whether your file review procedures address AI-related risks. A proportionate approach might include reviewing selected litigation files to see whether authorities were checked, whether delegated research was supervised and whether staff are following the AI policy. Weaknesses should lead to corrective action, further training or changes to procedures.

Client confidentiality: a second warning

The judgment also deals with confidentiality. In the first case, the adviser had put client emails and Home Office decision letters into ChatGPT. Paragraph 21 observed that doing so places the information in the public domain, and paragraph 60 states the point generally. Uploading confidential documents to an open-source AI tool such as ChatGPT breaches client confidentiality and waives legal privilege. It might itself warrant referral to the regulator and should in any event be referred to the Information Commissioner’s Office.

The Tribunal stated this in categorical terms. It also noted that closed tools which do not place information in the public domain, such as Microsoft Copilot, can be used for tasks like summarising without these risks. The SRA’s warning notice treats entering client information into a public AI tool as likely to breach confidentiality.

RPC’s August 2026 commentary notes that some commentators consider this part of the decision wrong in law. It also notes that it remains the only England and Wales authority on the point and that the SRA treats it as good law. The answer for any particular tool will turn on its technical arrangements, contractual protections and handling of uploaded data. Before approving any tool for client information, firms should establish how data is processed, retained, shared and possibly used for training. This belongs in information governance, not only IT procurement, and a confidentiality breach may also be a data protection matter.

What should a COLP do if fabricated case law is discovered?

The first priority is to establish the extent of the problem.

If a fabricated authority is found in work that has not yet been filed, prevent its use, verify the remaining research and examine whether the incident points to wider weaknesses in supervision or AI use.

If false authorities have already been submitted, the matter is more serious. Take advice promptly on the firm’s duties to the court, the need to correct the record and any procedural requirements. Consider whether the duty to report serious breaches to the SRA is engaged. In the first case before the Upper Tribunal, the adviser’s decision to report himself to his regulators was the reason the Tribunal did not make a referral itself.

Record the incident, investigate it and assess whether it reveals a systems or supervision failure. Check what else the same fee earner has worked on. In Munir, the Tribunal was troubled that the firm could not say how many other files the junior had touched. An isolated error and a pattern of unchecked research may need different responses, but neither should be ignored.

The wider lesson: responsibility cannot be delegated to AI

The Tribunal did not say solicitors must avoid AI. It said the use of legal AI programmes by properly trained professionals is a step forward in legal practice. Its warning concerns non-specialist AI, which requires rigorous checks on anything it produces. The problem arises when technology is allowed to bypass the checks that would apply to a junior colleague’s work. A solicitor would not accept an unverified authority because a trainee produced it, and the same applies when it comes from an AI system.

For managing partners this is a governance issue, for heads of department a supervision issue, and for COLPs a compliance risk to be managed through policy, training, monitoring and record-keeping.

The question every firm should ask is this: if a fabricated authority appeared in one of our court submissions tomorrow, could we demonstrate that reasonable steps had been taken to prevent it?

If the answer is uncertain, there is work to do.

How Enderley Consulting can help

At Enderley Consulting, we work with law firms on practical compliance arrangements that reflect both the regulatory requirements and day-to-day legal practice.

We can support firms through compliance audits and gap analysis covering AI use, supervision and verification procedures. Our COLP and COFA support services help compliance officers maintain effective controls and demonstrate a proactive approach to regulatory compliance. We also provide staff training through our Enderley Infohub practice management and training platform.

Contact Enderley Consulting to discuss how we can help strengthen your firm’s supervision and AI governance arrangements.

This article is for general information and does not constitute legal advice.