On 14 April 2026 the Civil Justice Council (CJC) closed its eight-week consultation on how legal representatives may use artificial intelligence when preparing court documents. The working group — chaired by Lord Justice Birss, Deputy Head of Civil Justice, with Mrs Justice Joanna Smith as deputy — is now drafting what would be the first formal rules in the civil courts of England and Wales on accountability for AI-generated material.
The proposals are narrower than the framing of "AI rules for lawyers" suggests, and that narrowness is a feature worth protecting.
What is actually on the table
Two changes do the real work. First, the rules governing trial witness statements would be amended so that any statement prepared with the involvement of a legal representative carries a declaration that AI was not used to generate its content — explicitly including altering, embellishing, strengthening, diluting or rephrasing the witness's own evidence. Second, the expert's declaration under Practice Direction 35 — which today requires experts to confirm their duty to the court and the truth of facts within their knowledge, with no mention of AI — would be amended so experts explain what use they made of AI beyond transcription and administrative tasks, and identify the tools they used.
The CJC's initial view is that no new rules are needed for statements of case or skeleton arguments produced with AI assistance, and that administrative uses such as transcription, spell-checking and the like require no declaration at all. That restraint matters: it signals that the working group is targeting a specific failure mode, not the technology.
Why now: the Ayinde warning
The backdrop is Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), handed down on 6 June 2025 by Dame Victoria Sharp, President of the King's Bench Division, sitting with Mr Justice Johnson. In the Ayinde judicial review, counsel put five non-existent cases before the court. In Al-Haroun, a witness statement relied on roughly 45 authorities, of which 18 simply did not exist. The court heard both under its Hamid jurisdiction — its inherent power to enforce the duties lawyers owe the court — and referred the barrister to the Bar Standards Board and the solicitor to the Solicitors Regulation Authority.
Dame Sharp was blunt about the tool: freely available generative AI tools "are not capable of conducting reliable legal research," and may "make confident assertions that are simply untrue" or "cite sources that do not exist." But she was equally clear that the duty is on the human. Anyone using AI for legal research "have a professional duty therefore to check the accuracy of such research by reference to authoritative sources, before using it." The court already holds a full toolbox for breaches: at paragraph 23 it lists public admonition, a costs order, a wasted costs order, striking out a case, referral to a regulator, the initiation of contempt proceedings, and referral to the police.
The strongest case for codifying
There is a genuine case for written declarations, and it should not be dismissed. Sharp herself found that existing guidance for judges, barristers and solicitors "is insufficient to address the misuse" of AI — a finding from the bench, not a press release. A signed declaration concentrates the mind at the moment of filing, creates a clean documentary record that makes later enforcement straightforward, and offers specific protection to a vulnerable category: witnesses whose own words could be silently rewritten by a model dressed up as their evidence. A bright-line rule is cheap, legible, and hard to argue with after the fact.
But the duty already binds — and over-reach is the real risk
The difficulty is that the Ayinde court needed no new rule to act. It enforced duties that already bind every advocate: a barrister's duty to the court and not to mislead it (Core Duty 1, Rule C3.1), and the SRA's parallel obligations on solicitors. The failure in these cases was not a missing declaration — it was a missing check. A lawyer who will file fabricated citations is unlikely to be stopped by a tick-box attesting they did not. There is a real danger that mandatory declarations become ritual: friction for the conscientious, no deterrent for the careless.
The two proposals are not equal. The witness-statement red line is sound, because generated evidence is categorically different from generated research — it goes to the integrity of testimony. The Practice Direction 35 model is the better general template: it does not ban a genuinely useful tool, it makes its use auditable. That distinction should hold the line. Extending mandatory declarations to every document filed would penalise exactly the routine, legitimate uses — research triage, summarisation, first-draft assistance — that the same judgment acknowledged. Sharp accepted AI "is likely to have a continuing and important role in the conduct of litigation in the future." The problem in Ayinde was unverified output reaching the court, not a lawyer who used software.
Get the proportionality right
The right settlement is the one the CJC has so far sketched: a firm declaration against AI-generated witness evidence, light-touch disclosure for experts, and explicit restraint on statements of case and skeleton arguments. What the courts should lean on hardest is the verification duty that already exists — and which Ayinde proves they will enforce, sanctions and all. New paper is only worth adding where it changes behaviour. Disclosure that makes AI use auditable does that; declarations that merely multiply attestations do not. England's civil courts have a rare chance to regulate a fast-moving tool without throttling it. The way to take it is to write less, and enforce what is already on the books.