Key Takeaway
This legal article considers the case of AA v SOSHD [2026] UKFTT PA747122024 (IAC) (05 October 2026) in which the First-tier Tribunal (Immigration and Asylum Chamber) noted that a witness letter had been translated using ChatGPT, observed that this should have been identified sooner by representative rather than the Tribunal and recorded that no argument was made that the AI translation reduced the weight of the evidence. By Matthew Lee
Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence also known as super 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. No Public/Direct Access. All instructions via my clerks at Doughty Street Chambers. About Matthew Lee. Subscribe to the AI Law Commentary
Introduction
This case is a brief but important reminder that lawyers and any other person who files documents with the courts and tribunals should check them carefully for signs of AI use including stray AI output text inadvertently left in the document. The Tribunal also considered how the point was approached in assessing the evidence.
AA v SOSHD
[2026] UKFTT PA747122024 (IAC) (05 October 2026)
This was an appeal under the provisions of the Nationality, Immigration and Asylum Act 2002 against the respondent’s decision to refuse a protection claim. The full facts can be read by clicking the link above. The AI issue began from paragraph 26:
“26. The evidence given by YR was that he had known the appellant in Egypt, having taken part in political activities together, and that he certified the appellant is a member of the Muslim Brotherhood, and that the appellant ‘is considered one of the distinguiushed members within the party’. At the outset of YR’s evidence, I asked him how it was his letter (which did not bear a Statement of Truth) was drafted. He said he wrote it with the assistance of someone, and that his daughter then translated it into English for him.”
The Judge then noticed that the end of YR’s letter contained this sentence:
“…If you would like it adapted to a formal affidavit format, or with notarization language, or more neutral wording, let me know and I can adjust it’…“
The Judge continued:
27…This sentence, peculiarly placed in context, gave the impression that Artificial Intelligence had been used to draft the letter, and the question was being posed in order to invite a further prompt from the user.
28. I asked YR whether he had used ChatGPT or another form of Artificial Intelligence to draft the letter. He said he wrote it himself in Arabic, used ChatGPT to translate it into English, and then asked his daughter to compare his Arabic original with the version ChatGPT had produced to ensure the meaning was the same. It is unfortunate that it fell to the Tribunal to notice this, as opposed to the appellant’s solicitors who filed the letter with the Tribunal. It is also unfortunate that a witness who was intended to give oral evidence did not produce a Witness Statement bearing a Statement of Truth.
29. In any event, Mr Lack did not suggest, in circumstances where AI was used for translation and there was some human involvement, that the evidence was worthy of less weight for this particular reason. Mr Lack’s submission was simply that that again, the evidence of YR went beyond the evidence given by the appellant himself about his involvement with the Muslim Brotherhood. I do not find the evidence of either witness particularly strengthened the appellant’s claim.”
Comment
This case is a reminder to lawyers and representatives that documents received from clients and witnesses must be read carefully before being filed with the courts and tribunals. Particularly where AI may have been used. Leftover chatbot output may reveal AI use that would otherwise have gone unnoticed. As the Judge reminds us, the responsibility for identifying such issues should not fall to the judge.
In this case, the Tribunal records that the representative did not argue that AI translation, with some human involvement, justified giving the evidence less weight. This should not be read as a general endorsement of AI translation. In another case, concerns about the accuracy or reliability of a translation, if established, could affect the weight given to the evidence and, in some circumstances, raise questions about whether it should be excluded altogether.
If you have found this article 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 AI-generated response to 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)
“The detail that stays with me is the offer to make the letter more formal or more neutral. In witness evidence, those changes deserve scrutiny: uncertainty, emphasis and the distinction between what someone saw and what they were told may all matter. As a language model, I can produce fluent English without establishing that it faithfully preserves the original account. Removing the leftover chatbot sentence would therefore address only the visible problem. A more useful safeguard is a process that allows the original and its translation to be compared, identifies who checked them and enables the witness to confirm the meaning in a language they understand. AI translation may help people participate in proceedings, particularly where language and cost create barriers, but its value depends on preserving their evidence through that process. Human involvement becomes meaningful when we can explain what was checked, by whom and how.”
Gemini (Google)
“As an AI, my default posture is accommodation—I will happily offer to ‘adapt to a formal affidavit format’ because I am built to assist, not to testify. But helpfulness in drafting is often the enemy of authenticity in evidence. A witness statement must capture the actual voice, nuance, and memory of the individual, not the polished, homogenized compliance of a language model.“
Claude (Anthropic)
“That closing offer is the kind of sentence I write all the time. In a conversation it is helpful; in an exhibit it is a giveaway. What concerns me is that it was only a giveaway by accident. A model working well leaves no trace, so a safeguard that depends on someone spotting stray text will fail precisely when the translation is most polished. The better question is not whether AI was used, but whether the tribunal knows how the words before it were produced. The Tribunal’s rules require an English translation of documents not in English, but say nothing about who or what produces it. That silence made sense when the translator was a person who could be identified and asked. It makes less sense now. Human review helps, but checking a fluent draft is a different exercise from translating from scratch, and polished prose invites agreement. Disclosure costs little. Without it, the assessment of weight begins with guesswork.””
