Part of the AI Regulation News hub.
The Upper Tribunal has rejected a suggestion put in cross-examination that a country expert's report was generated by AI, holding that there was no proper foundation for what it described as a serious allegation
Most AI-in-court reporting concerns lawyers caught filing fabricated citations. This is the other direction: a party alleging that an opponent's expert used generative AI, and a tribunal setting out why the indicators relied on did not support it.
Bottom line: A Country Guidance judgment in which a three-member Upper Tribunal panel rejected a suggestion, pursued in cross-examination, that a country expert's report had been generated in whole or in part by generative AI. The Tribunal found no proper foundation for the allegation and described it as a serious one.
Who this affects: Anyone instructing or challenging expert evidence, expert witnesses themselves, and advocates considering putting an AI-authorship allegation. The reasoning is not confined to immigration.
Issue date: Heard 9 to 13 March, 18 to 20 March and 27 July 2026. Issued 13 August 2026.
What changed: No rule changed. What the judgment supplies is worked reasoning on the evidential weight of the indicators commonly offered as signs of AI authorship, and a clear statement that alleging it is serious and requires a proper foundation.
Analysis: The most transferable part is the Tribunal's observation that two of the three indicators pointed the other way. An AI-written report would have been more likely to cite up-to-date material and to use idiomatic English, so an outdated omission and a non-idiomatic phrase are evidence against AI authorship rather than for it. That inverts how such points are usually deployed.
Primary sources: LR (Male VOTs) Albania CG, [2026] UKUT 00315 (IAC), Find Case Law
- Instrument
- LR (Male VOTs) Albania CG, judgment of the Upper Tribunal
- Citation
- [2026] UKUT 00315 (IAC)
- Court
- Upper Tribunal, Immigration and Asylum Chamber, heard at Field House
- Panel
- Deputy Chamber President Plimmer, Upper Tribunal Judge Blundell and Upper Tribunal Judge Landes
- Parties
- LR, subject to an anonymity order, and the Secretary of State for the Home Department
- Jurisdiction
- United Kingdom
- Status
- Issued 13 August 2026. Reported as a Country Guidance case.
- Bindingness
- The judgment is binding as Country Guidance on the country situation it addresses. The AI reasoning is not a rule of law but is the considered view of a three-member panel and is citable as such.
- Where the AI reasoning sits
- Paragraphs 155 to 159, within the Tribunal's assessment of the expert evidence of Professor Dalakoglou
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://caselaw.nationalarchives.gov.uk/ukut/iac/2026/315
What was alleged
The appellant relied on a country expert report from Professor Dalakoglou. For the Secretary of State, Mr Hansen pursued a suggestion in cross-examination that the report was not the Professor's own work and that it had been generated in whole or in part by generative AI.
The Tribunal recorded that the point was developed at length. It then dealt with it directly, and its first move was to classify it: it found no proper foundation for such a serious allegation.
That characterisation carries the weight here. An allegation that an expert did not write their own report goes to honesty and to compliance with the expert's duties, not merely to the quality of the opinion, and the Tribunal treated it accordingly.
The allegation rested on three bases: the repeated appearance of the same phrases within the report, the Professor's failure to mention the 2025 TIP report, and the appearance of the phrase to my opinion.
Why each indicator failed
On repetition, the Tribunal accepted the Professor's explanation that there was a degree of overlap and repetition in the questions he had been asked by the appellant's solicitors. Repetition in the answers followed from repetition in the instructions.
On the two remaining points the Tribunal went further, and this is the passage with the widest application. It held that the second and third points actually suggested the report was not written by AI, which would have been more likely to cite up-to-date material and use proper English.
So a missing recent source and a non-idiomatic phrase, both routinely offered as tells, were treated as evidence pointing away from machine authorship. The Professor explained that he had overlooked the more recent TIP report, and that although he is a British citizen, English is not his first language.
The Tribunal was pointed about the third indicator. It doubted that the question about the phrase to my opinion should even have been raised with an academic writing in his second language, and noted that he went on to use the same expression in his oral evidence.
The distinction the judgment preserves
The Tribunal did not treat the expert's evidence as beyond criticism. It held that the failure to consider the most recent TIP report would affect the weight given to the report.
But it separated that from authorship, holding in terms that the omission does not begin to suggest that he used AI. Weight and provenance are different questions, and a shortcoming going to the first is not evidence going to the second.
That separation is the practical lesson. A party who has a genuine point about currency or thoroughness can make it directly. Routing it through an AI-authorship allegation risks the point being answered as an allegation, and answered adversely, rather than being weighed on its own terms.
The Tribunal also rejected a related suggestion that the Professor was engaged in advocacy for the appellant, so the AI point did not stand alone. It formed part of a broader challenge to the expert that the panel found to be without merit in a number of respects.
What to take into practice
For an advocate, the judgment sets an implicit threshold. Alleging generative AI authorship is serious and needs a proper foundation, and a foundation built from stylistic impressions is unlikely to meet it, particularly where the impressions are equally consistent with an innocent explanation the witness can give on oath.
There is a second point that should be stated plainly, because the Tribunal did. Treating non-idiomatic English as a marker of machine authorship bears directly on witnesses writing in a second language. The Tribunal doubted the question should have been raised at all on that basis, and an advocate who builds an AI allegation on phrasing should expect that objection.
For an expert, the defensive lesson is procedural rather than stylistic. What answered the repetition point was an explanation of the instructions received. Keeping a clear record of the questions posed, the sources consulted and the drafting process is what makes such a challenge answerable.
For anyone tracking the field, note the asymmetry in the case law. The reported decisions overwhelmingly concern fabricated citations and undisclosed AI use. This is one of the comparatively few in which the allegation was made and failed, and it is useful precisely because a tribunal had to articulate what would and would not support one.
Three things are worth carrying from this judgment. First, the Upper Tribunal treated an allegation that an expert's report was AI-generated as a serious allegation requiring a proper foundation, not as an ordinary forensic point to be tried out in cross-examination. Second, it held that two of the three indicators relied on pointed the other way, because a report written by AI would have been more likely to cite up-to-date material and use proper English, which inverts the usual deployment of an outdated omission or an awkward phrase. Third, it kept weight and authorship separate: the failure to address the most recent TIP report affected the weight given to the report but did not begin to suggest AI use. The point about second-language English deserves particular attention, since the Tribunal doubted the question should have been raised at all with an academic writing in his second language. Practitioners should note that this is a Country Guidance decision, so it will be read widely.
Source File
https://caselaw.nationalarchives.gov.uk/ukut/iac/2026/315
Open the judgment on Find Case Law and confirm four things: the neutral citation [2026] UKUT 00315 (IAC) with the title LR (Male VOTs) Albania CG, issued 13 August 2026; the panel of Deputy Chamber President Plimmer, Upper Tribunal Judge Blundell and Upper Tribunal Judge Landes; paragraph 157, recording that Mr Hansen pursued a suggestion that the report had been generated in whole or in part by generative AI and that there was no proper foundation for such a serious allegation; and paragraph 158, setting out the three bases and holding that the second and third actually suggest the report was not written by AI.
We do not consider there to have been any proper foundation for such a serious allegation. Upper Tribunal, LR (Male VOTs) Albania CG, [2026] UKUT 00315 (IAC), paragraph 157
FAQ
What was the allegation?
That a country expert's report was not his own work and had been generated in whole or in part by generative AI. It was pursued in cross-examination on behalf of the Secretary of State.
How did the Tribunal deal with it?
It found no proper foundation for what it described as a serious allegation, and rejected each of the three bases on which it rested.
What were the three bases?
The repeated appearance of the same phrases in the report, the expert's failure to mention the 2025 TIP report, and his use of the phrase to my opinion.
Why did two of them point the other way?
The Tribunal held that a report written by AI would have been more likely to cite up-to-date material and to use proper English. An outdated omission and a non-idiomatic phrase therefore suggested human authorship rather than machine authorship.
Did the expert's evidence escape criticism?
No. The Tribunal held that the failure to consider the most recent TIP report would affect the weight given to the report. But it held that this did not begin to suggest AI use, keeping weight and authorship separate.
Does this bind other courts?
The judgment is Country Guidance on the country situation it addresses. The reasoning about the AI allegation is not a rule of law, but it is the considered view of a three-member Upper Tribunal panel and can be cited as persuasive.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.