Three regulatory bodies have published guidance on AI in legal proceedings since October 2025. Together, they create a new framework where verification of AI-generated legal research is non-negotiable — and the choice of research tool becomes a compliance decision.
In June 2025, a barrister and solicitor submitted five fabricated cases to the High Court in a judicial review. One purported to be from the Court of Appeal. When asked to produce copies of the authorities, they could not — because the cases did not exist. The judge called it “wholly improper” and rejected the parties’ characterisation of the fabrications as “minor citation errors” or “cosmetic errors.”
The case — Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin) — was not an isolated incident. It was the case that forced the English judiciary, the Bar Council, and the Civil Justice Council to formalise what had previously been informal concern: generative AI is producing fictitious legal authorities, and practitioners are submitting them to courts without verification.
Since then, three separate regulatory bodies have published guidance. Together, they establish a new framework governing how AI-generated legal research reaches the courtroom in England and Wales. Every practitioner — and every legal research tool — now operates within that framework.
The Judicial Guidance: October 2025
The judiciary published updated AI guidance for all judicial office holders in October 2025, replacing an earlier version from April of the same year. The guidance applies to judges, tribunal members, clerks, judicial assistants, legal advisers, and support staff.
The core position is permissive but cautious: judges are not prohibited from using AI, but must take full personal responsibility for whatever goes out in their name. A judge who uses AI should only use a system they are confident is secure.
On hallucinations, the guidance is direct. AI tools may fabricate cases, citations, or quotes. They may refer to legislation, articles, or legal texts that do not exist. They may provide incorrect or misleading information regarding the law or how it might apply. All AI-generated outputs must be independently verified before use.
The guidance also addresses the use of AI by legal representatives appearing before the courts: all representatives are responsible for the material they put before the court and have a professional obligation to ensure it is accurate and appropriate. The judiciary does not currently require legal representatives to disclose their use of AI in preparing submissions — provided the AI is used responsibly. But that position is conditional on context, and may change.
The Bar Council Guidance: November 2025
The Bar Council updated its guidance on generative AI in November 2025. The update is significant for two reasons: it now covers legal-specific AI tools (not just general-purpose models), and it introduces a section on mandatory verification.
Previous versions of the guidance focused primarily on general-purpose systems like ChatGPT. The updated guidance explicitly references legal-specific tools including Lexis+ AI, Clio Duo, and Thomson Reuters Co-Counsel. This matters because practitioners may have assumed that legal-specific tools were inherently more reliable than general-purpose ones. The Bar Council draws no such distinction — all AI outputs require verification regardless of the tool’s marketing claims.
The new section on mandatory verification and human oversight states that AI may assist efficiency but cannot replace independent research, analysis, and professional judgment. This is the Bar Council’s clearest statement yet that the duty to verify is absolute and non-delegable.
The professional consequences are explicitly stated: misleading the court via AI-generated material — even inadvertently — may amount to incompetence and serious professional misconduct, engaging Core Duties 1, 3, and 5. The word “inadvertently” carries significant weight. It means that failing to verify an AI-generated citation is not excused by ignorance of the fabrication. The duty runs to the output, not the intent.
The Civil Justice Council Consultation: February 2026
On 17 February 2026, the Civil Justice Council launched an eight-week consultation on whether rules are needed to govern the use of AI by legal representatives in preparing court documents. The working group is chaired by Lord Justice Birss, Deputy Head of Civil Justice, with Mrs Justice Joanna Smith as Deputy Chair.
The consultation makes different proposals for different types of documents — a distinction that reflects a sophisticated understanding of where AI risk is highest.
For statements of case, skeleton arguments, and other advocacy documents, the CJC proposes no new rules — provided they bear the name of the legal representative taking professional responsibility. The existing professional duty to verify is considered sufficient.
For witness statements, the proposal is more restrictive. The CJC has proposed that solicitors and barristers should be required to confirm they did not use AI to generate the content of witness statements intended for trial. This reflects the unique evidentiary function of witness statements — they are supposed to represent a witness’s own recollection, not an AI’s synthesis of what they might have said.
For disclosure, the CJC sees no pressing need for AI-specific rules at this stage.
No mandatory disclosure or certification requirements are currently in force. Any new rules would require amendments to the Civil Procedure Rules 1998. But the direction of travel is clear: the regulatory framework is tightening, not loosening.
What Ayinde Established
The Ayinde judgment deserves specific attention because it is the leading English authority on AI-fabricated citations in court proceedings.
The combined judgment — which also addressed Al-Haroun v Qatar National Bank — dealt with two separate instances where generative AI was suspected to have produced fictitious case law that was then relied upon in litigation. In Ayinde, five fake cases were included in the statement of facts and grounds for a judicial review concerning social housing obligations under the Housing Act 1996.
The court’s response was unambiguous. Fabricated cases in a pleading are not citation errors. They are not cosmetic. They are, in the court’s words, “wholly improper.” The judgment reminded all legal representatives of their fundamental professional duty to take responsibility for their own submissions — including by taking steps to verify AI output.
The practical implication is that “I didn’t know the AI made it up” is not a defence. The duty to verify existed before Ayinde; the case simply confirmed that it applies with full force to AI-generated material and that breach of that duty will be treated seriously.
What This Means for Legal Research
The combined effect of these three regulatory interventions is a framework built around one principle: verification is non-negotiable.
The judiciary requires judges to independently verify all AI outputs. The Bar Council requires barristers to verify all AI outputs, on pain of professional misconduct. The CJC is consulting on whether to formalise verification requirements in the procedural rules themselves.
For practitioners, this creates a practical problem. Verification takes time. Checking whether a case exists, whether the citation is correct, whether the quoted passage actually appears at the cited paragraph, whether the case has been subsequently overruled — these are the tasks that AI was supposed to accelerate, not complicate. If every AI-generated authority requires manual verification against primary sources, the efficiency gains diminish rapidly.
This is where the choice of research tool becomes a regulatory compliance decision, not just a productivity preference. A tool that generates plausible-looking citations and leaves verification to the user creates professional risk. A tool that verifies every citation against official databases before the output reaches the practitioner reduces that risk structurally.
The distinction matters because the regulatory framework does not distinguish between AI-generated errors that the practitioner should have caught and errors that were caught by the tool before delivery. What reaches the court is what counts.
The Verification Standard Going Forward
The trajectory is clear from the regulatory activity in the past twelve months. The judiciary published guidance. The Bar Council tightened its requirements. The CJC is consulting on procedural rules. Each intervention has been more specific and more demanding than the last.
Practitioners who adopt AI-assisted legal research need to ask one question of any tool they use: does this tool verify its outputs against primary sources before I see them, or does it leave that to me?
The answer determines whether the tool helps you meet the emerging regulatory standard — or becomes the reason you fall short of it.
Search the decisions where courts have addressed AI-generated content and citation verification.
Search the Law verifies every citation against The National Archives and legislation.gov.uk before it reaches your document. No fabricated cases. No phantom authorities. Every citation linked to its official source.
Try it free at searchthe.law
Sources and further reading:
- Judiciary of England and Wales, “Artificial Intelligence (AI) — Judicial Guidance” (October 2025)
- Bar Council of England and Wales, “Updated Guidance on Generative AI for the Bar” (November 2025)
- Civil Justice Council, “Use of AI for Preparing Court Documents: Interim Report and Consultation” (February 2026)
- R (Ayinde) v London Borough of Haringey; Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin)
- SRA, “Compliance Tips for Solicitors Regarding the Use of AI and Technology”