AI in courts demands disciplined legal workflows

The English courts are now dealing with a steady flow of AI-generated legal material. Some is produced by litigants in person, and some is filed by lawyers. The most visible problem is the easiest to understand: false cases, invented quotations and confident legal propositions that fall apart when checked.

That is a serious professional issue. In Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), the Divisional Court made clear that putting false legal material before the court can have serious consequences, including referral to a professional regulator and wasted costs, and it may amount to contempt of court. The core message is that lawyers remain responsible for the material they put before the court. That is also the starting point for the Civil Justice Council’s consultation on AI and court documents: the proposed procedural response is relatively targeted, with particular focus on witness statements and expert evidence, rather than a wholesale new rulebook for every AI-assisted litigation task.

Moving beyond hallucinations

We should not cite hallucinated cases. But if the profession treats AI risk as no more than “do not cite fake cases”, the debate becomes too narrow. It reduces responsible AI use to a few familiar slogans: keep a human in the loop, check everything, never trust the machine. Those points are correct, as far as they go. But they are insufficient, and they also risk obscuring the positive case for AI in legal work. 

Used well, AI can improve the quality of legal analysis, not just make it faster. It can help lawyers interrogate source material, test arguments, spot gaps, structure advice and work more effectively with large bodies of information. 

The current judicial guidance on AI, issued in October 2025, still cautions against using AI tools to conduct legal research. But it already needs to be read alongside how quickly practical use cases are developing. One of the most constructive themes of the recent City of London Law Society discussion on AI, privilege and the courts was Sir Colin Birss’s pragmatic and positive keynote. He explained that the judiciary is now using secure AI systems for practical tasks such as transcription, anonymisation, checking draft judgments for internal inconsistency and finding material in emails. In short, AI can be incredibly effective for legal tasks, if used responsibly.

That means looking at tool choice, source material, prompting, verification, supervision, confidentiality and privilege. The problem is about more than bad legal outputs; it is about bad workflows.

From checking outputs to developing better workflows

The core control is still verification. Any authority, quotation or legal proposition generated by AI needs to be checked against the original source. That applies whether the tool is a public chatbot, an enterprise system or a specialist legal product. 

But responsible use starts earlier than the final check. It starts with your choice of model. Reasoning or “thinking” models are materially better suited to legal tasks than faster, lighter models. They are much less likely to hallucinate, and substantially less prone to providing confident but unsupported answers. Where possible, choosing a reasoning model is an important starting point in legal work for accuracy and reducing hallucination risk. That does not remove the need to verify the answer against the underlying sources, but it reduces avoidable risk before the final check begins.

Solicitors need to think carefully about the task they are giving the tool. A broad prompt asking for “the law on X” is much more vulnerable to error than a prompt anchored in identified source material, a defined jurisdiction, a clear factual context and a specific output format. Where possible, AI should be connected to reliable legal databases or used inside specialist legal tools that can retrieve and cite source material. Even then, the solicitor needs to test whether the answer actually follows from the sources.

This is where training and supervision become central. Junior lawyers, paralegals and trainees may be among the most frequent users of AI tools, because they often handle first-pass research, summaries, chronologies and draft notes. 

The recent Upper Tribunal decision in UK v Secretary of State for the Home Department [2026] UKUT 00081 (IAC) makes the supervision point directly: “[t]he qualified legal professional with conduct of the matter is expected to ensure that... documents are checked, that errors are identified, and that only accurate documents are sent to the tribunal.” The Tribunal was also critical of the suggestion that a firm had no mechanism for staff to use AI, noting that this overlooked the fact that “anyone with access to Google has access to AI." 

That is a useful warning about shadow AI. Firms cannot assume that AI is irrelevant because they have not formally rolled out a tool. They need to supervise use and set practical rules before unapproved use becomes the default. 

The Divisional Court in Ayinde made this point too, emphasising that “practical and effective measures must now be taken by those within the legal profession with individual leadership responsibilities (such as heads of chambers and managing partners) and by those with the responsibility for regulating the provision of legal services.” This is about training, with the Court stating that these “measures must ensure that every individual currently providing legal services within this jurisdiction … understands and complies with their professional and ethical obligations and their duties to the court if using artificial intelligence.” 

These cases are constructive because they move the discussion beyond blame after the event. Firms need more than an AI policy that tells people what not to do. They need practical guidance on good use cases, bad use cases, and best practice. 

For barristers, the BSB's recently published May 2026 guidance takes the same practical approach: responsible AI use is treated as a competence and practice-management issue, requiring careful tool choice, confidentiality controls, verification, appropriate records and clear rules for use. 

The Law Society’s response to the CJC consultation makes a similar point in regulatory terms: it calls for SRA guidance on baseline AI competence, verification, training and firm-level governance, and cautions “against an approach that relies primarily on disclosure obligations, without equivalent focus on the capability and governance required to make those obligations meaningful in practice”.

Public tools and private tools

Much of the legal profession’s AI debate has focused on public and private tools, as reflected in the judicial and Bar guidance on the use of AI. The caution is sound: confidential or privileged client material should not be put into public tools where the terms, retention settings, access rights or training use are unclear. This is important not least because confidentiality is the gateway to privilege; a secure, private tool is much more likely to provide the level of confidentiality on which any privilege claim depends.

But the notion of 'public bad, private good' is still too crude. A secure tool can still be used badly: it may create unnecessary records, auto-save prompts, expose outputs through enterprise search, or encourage speculative drafts that are difficult to explain.

Security is only one part of it. You also need to look at the task, the input material, the records created, who can see them, and how the output is checked and used.

Practical consequences

For solicitors, the practical message should be positive: use AI, but use it well.

Legal teams should decide which tools are approved for which tasks, train lawyers on effective prompting and verification, and treat AI-generated materials as part of the matter record. Where prompts or outputs relate to advice, they should be kept in controlled legal workspaces, with circulation limited to those who need to see them. 

Clients also need guidance. One emerging risk is that clients will increasingly arrive having already asked a public AI tool whether they have a legal problem, what their position is, or how they should respond. That may create unprivileged material before the legal team is involved. Businesses should be told to direct legal questions to legal, to avoid using public tools for confidential issues, and, where AI is used to prepare material for lawyers, to keep it clearly linked to the purpose of seeking legal advice. 

The hallucination cases are a necessary warning. But they should not define the whole professional conversation. Responsible AI use for solicitors is about avoiding fake authorities, but it is also about developing legal workflows that harness the benefits of AI for the benefit of clients.

Andrew Holland, Lydia Savill and Antonia Croke of Hogan Lovells also contributed to this article.