Key Takeaway
This legal article examines the recent decision in Abraham v Hound Technology Ltd from the Employment Tribunal. The article draws out practical lessons for lawyers, litigants in person and the justice system from that decision. By Matthew Lee.
Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI generated documents and evidence in court Subscribe to the AI Law Commentary here.
Publication date: 30 June 2026
Introduction
This is another brief article, or at least what my idea of brief is, concerning a recent decision from the Employment Tribunal. The judgment in Abraham v Hound Technology Ltd is short, but it captures several themes that seem to be increasingly important across courts and tribunals. Those include the speed with which AI can produce legal documents, the risk that speed may overtake judgment, the confidentiality implications of using public AI tools and the legitimate role these tools may play for litigants with dyslexia or other relevant needs.
I am not concerned here with the merits of the underlying employment claim or the interim relief application. The point of interest for me is how AI may have been utilised in this case and how the Judge responded. The observations in this case go beyond the Employment context.
Abraham v Hound Technology Ltd
On 31 March 2026, the Employment Tribunal refused the Claimant’s application for interim relief and the Claimant stated an intention to appeal. Within minutes of the hearing concluding, the Claimant submitted an application for reconsideration, accompanied by a four-page statement of grounds in support of the application. Approximately 10 minutes later the Claimant submitted amended grounds, to which had been added a further, fifth ground. Written reasons were also requested.
On 1 April 2026, the Respondent responded to the reconsideration application and made the following observations:
“We refer to the Claimant’s applications for reconsideration. We respectfully note that the first of those applications was lodged at 1:06pm on 31 March 2026, 10 minutes after Employment Judge Tynan had delivered his oral Judgment. The Claimant’s first application for reconsideration is detailed and is 4 pages long. The Claimant later lodged a second application for re-consideration, that was 5 pages long, at 1:20 pm.
Given the detail in these documents and the very limited time in which they were produced, we consider it inconceivable that the applications were not prepared by AI. We are concerned what information was inserted into the AI software, in order to prepare an application for a reconsideration of a Judgment that had only just been handed down orally and in respect of which it would have been very difficult to prepare a verbatim note. The Claimant has assured us that she did not record yesterday’s hearing but used her contemporaneous notes to prepare the applications” [para 3]
Employment Judge Tynan then invited the Claimant to provide her comments. In doing so, he said:
“…I should stress that I have not formed any view as to whether the Claimant has behaved improperly, in particular by recording and/or generating a transcript of the hearing without the Tribunal’s permission. As many unrepresented parties now do, the Claimant readily accepts that she has relied upon AI in these proceedings. During the hearing, she disclosed that she has dyslexia and explained that she had used AI to finesse her documents, but that the content was her own. In that regard, one of our primary concerns as judges is to ensure that AI generated documents reflect rather than suggest a party’s case and evidence.” [para 4]
The Judge then referred to the Claimant’s comments in response to his direction. He recorded that she had said she began preparing the reconsideration application while he was giving his decision. The Judge then added:
“…Particularly given her processing difficulties as a result of her dyslexia, it raises a question in my mind as to how the Claimant was able to focus on what I was saying and draft or edit her reconsideration application at the same time. Be that as it may, I consider that I can decide the reconsideration application without offering any further view in the matter, including how it is that the Claimant was able to submit her application, grounds and amended grounds all within approximately twenty minutes of the hearing concluding.” [para 5]
The Judge then moved on to consider the application, finding that there was no reasonable prospect of the decision being varied or revoked.
Conclusion
This is a very brief judgment, but a close reading raises several interesting points which, I think, will assist everyone, no matter their area of law or jurisdiction, as to what is happening as AI use becomes more common in courts and tribunals.
Firstly, we really are entering a new era of legal work when, 10 minutes after an oral judgment is delivered, a four-page detailed application is created followed shortly by an amended version. The question raised by the Judge is an important one. Even if AI can help produce a document very quickly, has there been enough time to digest what the judge has said to form a reasoned view on whether there is merit in applying for reconsideration or appealing and to decide which points should properly be accepted or challenged?
We don’t know exactly what happened in this case and, in certain cases, there may be good reasons for appealing or applying for reconsideration quickly. But reflection matters. Lawyers often need time to read, think, compare the decision with the evidence and law and distinguish between disappointment with an outcome and an arguable error. AI may help with expression, structure and speed. But speed is not always the same as judgment. The risk is that the very feature that makes AI attractive may sometimes work against the party using it.
This is, in a sense, a significant problem with AI. The very fact that it allows litigants to act this quickly could actually harm their case.
The second point is confidentiality. It has been clear for some time that everyone must be careful about what they put into any AI system. The judicial guidance on AI makes this point in firm terms:
“II. Uphold confidentiality and privacy
Do not enter any information into a public AI chatbot that is not already in the public domain. Do not enter information which is private or confidential. Any information that you input into a public AI chatbot should be seen as being published to all the world. The current publicly available AI chatbots remember every question that you ask them, as well as any other information you put into them. That information is then available to be used to respond to queries from other users. As a result, anything you type into it could become publicly known. You should disable the chat history in public AI chatbots if this option is available, as it should prevent your data from being used to train the chatbot and after 30 days, the conversations will be permanently deleted. This option is currently available in ChatGPT and Google Gemini but not in some other chatbots. Even with history turned off, though, it should be assumed that data entered is being disclosed. Be aware that some AI platforms, particularly if used as an App on a smartphone, may request various permissions which give them access to information on your device. In those circumstances you should refuse all such permissions. In the event of unintentional disclosure of confidential or private information you should contact your leadership judge and the Judicial Office. If the disclosed information includes personal data, the disclosure should be reported as a data incident. Details of how to report a data incident to Judicial Office can be found at this link: Data breach notification form judiciary. You should treat all public AI tools as being capable of making public anything entered into them.”
The guidance is addressed to judicial office holders and related court and tribunal staff, but the underlying confidentiality and privacy points are plainly relevant to court users and lawyers too. The Respondent’s stated concern about “what information was inserted into the AI software” is a reminder that a party who places case material into an AI tool may run into difficulties, depending on the tool used, the terms on which it is used, and whether the material is private, confidential or privileged. Readers should also consider the observations in UK and R (on the application of Munir) v Secretary of State for the Home Department [2026] UKUT 81 (IAC) discussed here
Thirdly, the Tribunal expressly formed no view as to whether the Claimant behaved improperly by recording and/or generating a transcript of the hearing without the Tribunal’s permission. The judgment records the Respondent’s statement that the Claimant had assured it that she did not record the hearing and had used her contemporaneous notes. There is not enough information in the judgment to comment further on that case-specific point, but the general issue may require clearer guidance given the ease with which AI tools can now record, transcribe and summarise hearings.
Fourthly, there is the use of AI to assist people with dyslexia. I have said several times before that AI can be a valuable support for people with specific learning difficulties and I have commented on a number of cases from the UK and Australia addressing this issue. It is important not to lose sight of that. Used carefully, these tools may help people organise their thoughts, express themselves more clearly, and participate more effectively in proceedings. That is a significant and proper use of AI, even as this same case raises harder questions about speed and authorship.
The fifth point is, to my mind, the most important takeaway. The judge observed that one of the primary concerns of judges is to ensure that AI-generated documents “reflect rather than suggest a party’s case and evidence”. That is a carefully expressed and important distinction.
I have written about this before, in the context of witness statements, in my article on The Dangers of Using AI to Assist with Witness Statements: Lessons from Godwin v Godwin. In that case, the court was faced with witness statements that the Defendant and his witness said had been prepared with limited assistance from a digital assistant. The Judge was not satisfied. He could see no good reason for sophisticated litigants to have used ChatGPT at all when the spelling and grammar tools in a word processor were available and could not be sure the statements were in the witnesses own words.
That concern is close to the one identified by Employment Judge Tynan, although it arises in a different procedural context. In Godwin, the issue was whether the witness statement remained the witness’s own evidence, or whether AI had altered, supplied or reshaped it in some material way. In Abraham v Hound Technology Ltd, the issue was whether AI-assisted reconsideration grounds reflected the litigant’s own case and evidence, rather than supplying or reshaping it. The principle is similar, but the evidential and procedural setting is different.
My own concern remains a very practical one in this regard. I am increasingly unconvinced that current technology, or human assessment, can reliably tell whether AI has been used in a legal document and, if so, to what extent. Absent an admission, a clear audit trail, or some obvious feature in the document, how are courts and opponents realistically to identify improper or excessive AI use? That is not a reason to abandon the principle that a party’s case and evidence should be their own. It is a reason to frame any future rules or guidance carefully, so that they can be evidenced and applied fairly without generating disproportionate satellite litigation or unfair allegations. The Civil Justice Council has already consulted on the use of AI for preparing court documents and I look forward to reading the final report.
If you have found this helpful, you may wish to subscribe to my AI law articles via the Substack below. Please do continue sending me decisions that you think should be tracked, whatever the jurisdiction. One of the clearest lessons from this work is that lawyers, judges and court users in different systems are often confronting closely related problems. Sharing decisions, experiences and reflections across practice areas and across jurisdictions remains one of the most constructive ways to respond.
Final Words from Popular LLMs
This is the point at which I invite an LLM to provide an independent comment on the preceding article. At times the observations are slightly off the mark, at other times they are perceptive, and occasionally they produce something entirely unexpected. Whatever the outcome, I find it a useful and revealing exercise to see what emerges.
ChatGPT (OpenAI)
“Tools that make submissions almost instantaneous can undermine the reflective function that reconsideration and appeal decisions require. Courts should not treat speed, polish, or AI use as suspicious in itself, especially where accessibility needs are in play. But the boundary should be firm: AI may assist formulation, organisation and accessibility, but it should not originate evidence, invent grounds, or obscure who is actually advancing the case. The practical answer is not blanket disclosure of every use of AI, which risks satellite disputes and disadvantages litigants in person, but targeted obligations where evidence, confidentiality, privilege or legal authority are affected, with judges retaining power to ask direct questions when timing, content or context gives a real reason to do so.”
Gemini (Google)
“relying on public large language models (LLMs) to process unverified contemporaneous notes of oral judgments invites serious confidentiality breaches and the risk of hallucinated arguments. Courts shouldn’t shut the door on AI — especially given its genuine benefits for accessibility and processing difficulties — but they must establish strict procedural boundaries requiring parties to certify that AI-generated submissions accurately reflect the facts of the case, rather than what an algorithm predicts sounds persuasive.“
Claude (Anthropic)
“The harder problem is the “reflect rather than suggest” line itself. It sounds principled, but it has no obvious limit: a barrister who settles grounds of appeal also supplies framing and arguments the client never independently formed, and nobody calls that authorship by the lawyer. Unless the distinction is tied to something concrete, whether the litigant adopts the content and can speak to it, whether the factual assertions are her own, it risks shading into a general unease about unrepresented parties who use AI, pressed harder against them than against the represented opponent who has the same tool and a professional shield in front of it. Employment Judge Tynan was right to dispose of the application without resolving any of this, and the article is right that detection is the real constraint. But the candid position is that “authorship” is carrying more weight than it can yet bear, and any rule built on it will be unenforceable until someone says what reflecting, rather than suggesting, actually requires.“
