Tuesday, 4 August 2026
The growing use of artificial intelligence in legal proceedings is forcing courts, regulators and professional bodies in the United Kingdom to reconsider how existing procedural rules should apply to AI-assisted documents.
A recent High Court judgment has highlighted the particular risks surrounding witness statements, while a wider consultation by the Civil Justice Council is examining whether new disclosure and accountability requirements are needed for court documents ranging from pleadings to expert reports.
The central challenge is not whether AI can assist legal work, but how courts can ensure that evidence remains authentic, legal submissions are accurate and responsibility for the finished document is clear.
High Court questions AI-assisted witness statements
In Godwin v Godwin [2026] EWHC 923 (Ch), the High Court considered a family dispute concerning arrangements for the burial of the deceased.
The case attracted wider attention because two witnesses acknowledged using ChatGPT while preparing their trial statements. They described the software as a digital assistant used for grammar, spelling and presentation, while maintaining that the original wording had not been altered.
However, the Court was not shown the initial drafts and therefore could not verify the extent of the AI intervention.
His Honour Judge Klein found that the explanations provided did not remove uncertainty over whether the statements remained entirely in the witnesses’ own language. The evidence was consequently treated with caution.
The judgment reinforces the requirements of Practice Direction 57AC and Practice Direction 32, under which factual witness evidence must reflect the witness’s own recollection and wording rather than language created or reshaped by another party or system.
The case does not establish a general prohibition on AI, but it demonstrates the evidential difficulty created when the Court cannot determine precisely how a tool was used.
Proposed declaration for witness evidence
The Civil Justice Council has been considering similar concerns through its consultation on AI-assisted court documents.
Its interim position indicates support for an additional declaration confirming that AI has not been used to generate the substance of a trial witness statement. That restriction could cover not only creating new material but also rewording, strengthening, softening or embellishing a witness’s account.
A comparable requirement has been proposed for witness statements governed by the broader rules under CPR Part 32.
Both the Law Society and the Bar Council have supported stronger safeguards in this area. Their concern is that even apparently limited editing tools may make suggestions or restructure language in a way that moves the statement away from the witness’s own expression.
Until the meaning of AI-generated content is defined more precisely, the lowest-risk approach for legal representatives is to avoid using generative tools in the preparation of witness evidence.
Different treatment for legal submissions
The emerging regulatory approach is less restrictive for pleadings, skeleton arguments and other advocacy documents.
The Civil Justice Council’s preliminary view is that additional procedural rules may not be required where the document identifies the lawyer who accepts professional responsibility for its contents.
This reflects the fact that legal submissions are already drafted on behalf of clients and are not required to reproduce a person’s direct memory in the same way as a witness statement.
Nevertheless, professional responsibility remains unchanged. Lawyers must verify all authorities, quotations and factual assertions, regardless of whether AI was involved in producing the text.
The Law Society has favoured greater transparency over AI use, while the Bar Council has generally supported reliance on existing responsibility and verification obligations for advocacy documents.
AI may improve access to justice
Judicial concerns over accuracy have been accompanied by recognition that AI can also make court materials easier to understand.
During evidence to the House of Lords Constitution Committee, Lady Chief Justice Baroness Carr said AI-assisted submissions from self-represented litigants can sometimes be clearer and more useful to judges than documents prepared without such support.
This is particularly relevant where individuals do not have access to legal representation.
However, courts have also made clear that litigants in person remain responsible for false cases, invented quotations or fabricated citations submitted in their name. Their regulatory position may differ from that of a solicitor or barrister, but inaccurate material is no more acceptable because it was produced using AI.
Expert reports may require disclosure
Expert evidence presents a different regulatory question because AI may form a legitimate part of technical analysis.
Proposals under consideration would require experts to identify non-administrative uses of AI and name the tools used when preparing their reports. Routine transcription and similar support functions may be treated differently from systems used to analyse evidence or generate substantive conclusions.
The Law Society has supported a disclosure requirement. The Bar Council has warned, however, that experts will need clear guidance on how much detail must be provided.
Any future rule will need to distinguish between AI that assists an expert and AI that substitutes its own analysis for the expert’s independent judgment.
First authorised AI law firm tests the boundaries
The debate is also developing through new models of legal practice.
Garfield AI, described as the first fully AI-based firm authorised by the Solicitors Regulation Authority, secured a successful outcome in a small-claims case earlier in 2026. The platform carried out much of the preparatory work, including the case bundle and witness-statement process, while a barrister conducted the courtroom advocacy.
Its model is limited to claims of up to £10,000 and operates under a specific regulatory authorisation. Small-claims proceedings are also subject to less demanding requirements than commercial trial witness statements governed by Practice Directions 57AC and 32.
The case therefore demonstrates that regulated AI-led legal services are possible, but not that the same methods can automatically be transferred into more complex litigation.
Courts warn against AI witness coaching
Judicial concern extends beyond document preparation.
In a recent criminal appeal, the Court of Appeal warned witnesses against using AI to rehearse or prepare their testimony. The Court emphasised that witness coaching is prohibited regardless of whether it is carried out by a person or a technological system.
AI-based preparation could influence how a witness recalls or presents evidence and, in serious cases, could create sufficient unfairness for the Court to consider excluding the evidence.
The same concern is likely to influence civil proceedings, where the integrity and independence of witness recollection are equally important.
Hallucinated authorities remain a major risk
Recent proceedings in the UK and United States have continued to expose the danger of fabricated legal content.
British cases including R (Ayinde) v Haringey LBC and Taiwo v Homelets of Bath Ltd have addressed false authorities produced through AI systems. Other incidents have involved law firms submitting documents containing invented citations and then using AI again when attempting to explain the original mistake.
The recurring nature of these errors suggests that professional verification controls have not always kept pace with the rapid adoption of generative tools.
For insurers and law firms, the exposure includes wasted costs, professional disciplinary action, reputational damage and potential negligence claims.
Confidentiality and privilege concerns
A further risk arises when confidential documents are uploaded into publicly accessible AI systems.
In UK v Secretary of State for the Home Department [2026] UKUT 00081 (IAC), the Upper Tribunal warned that placing confidential material into an open-source AI tool could amount to publishing it into the public domain, potentially breaching confidentiality and waiving legal professional privilege.
The precise legal consequences will depend on the system, its contractual terms and how submitted data are processed. However, the decision underlines the need for firms and clients to understand where information is stored, whether it is retained and whether it may be used to train or improve the service.
Regulation is moving, but responsibility remains unchanged
The Civil Justice Council’s final recommendations are expected later in 2026 and may lead to amendments to civil procedure rules and practice directions.
The likely outcome is not a single rule applying to every form of AI use. Witness statements, legal arguments and expert evidence serve different purposes and will probably require different standards.
What is already clear is that AI does not transfer responsibility away from the person signing, submitting or relying upon a document. Lawyers, witnesses and experts must still be able to explain how material was prepared, confirm its accuracy and demonstrate that it complies with the applicable procedural rules.
Until clearer standards are introduced, the safest distinction is between technology that supports administrative work and technology that changes the substance of evidence or professional judgment.
Source: CMS
