A tribunal has ruled on what does not prove an expert used AI to write his report

Tech and AI

A tribunal has ruled on what does not prove an expert used AI to write his report

By Staff Writer  |  16 August 2026

The Gothic stone frontage, twin spires and rose window of the Royal Courts of Justice under a grey sky

The allegation was put in cross-examination and rested on three things: repeated phrases, an omitted source and one odd turn of expression. A three judge panel found no proper foundation for any of it, and held that two of the three pointed the other way.

A decision issued on Thursday 13 August by the Upper Tribunal, Immigration and Asylum Chamber, takes on a question that will be put in a great many hearings over the next few years: what does it take to show that a written expert report was produced by a machine.

The allegation arose in cross-examination of one of the appellant's expert witnesses, Professor Dalakoglou, a Full Professor of Social Anthropology and Sociology of Development at Vrije Universiteit Amsterdam. Counsel for the Secretary of State put to him that the report was not his own work and had been generated in whole or in part by generative artificial intelligence.

The panel of Deputy Chamber President Plimmer, Upper Tribunal Judge Blundell and Upper Tribunal Judge Landes disposed of it in three paragraphs.

We do not consider there to have been any proper foundation for such a serious allegation

Deputy Chamber President Plimmer, Upper Tribunal Judge Blundell and Upper Tribunal Judge Landes, Upper Tribunal (Immigration and Asylum Chamber)

The three indicators, and what the panel made of them

The allegation was built on three observations about the document. The same phrases appeared repeatedly within it. The expert had failed to mention a trafficking in persons report published in 2025. And the report used the phrase "to my opinion".

On the first, the tribunal accepted that there was a degree of repetition, and accepted the expert's explanation for it: he had been asked a set of questions by those instructing him which themselves overlapped, and he had answered them.

The second and third are where the reasoning turns, because the panel held that both of them cut against the allegation rather than supporting it. A report written by a machine, the tribunal said, would have been more likely to cite up-to-date material and use proper English. The omission of the more recent source and the slightly off idiom are what a human author produces, not what a language model produces. The expert explained that he had overlooked the newer report, and that although he is a British citizen English is not his first language.

The omission was still held against the report, but on weight rather than on authorship. The tribunal said the failure to consider the most recent material would affect the weight given to the report, and in the same breath that it did not begin to suggest AI use.

On the third point the panel went further and questioned whether it should have been raised at all, saying it very much doubted the question should even have been put to an academic writing in his second language, and noting that he had used the same expression in his oral evidence.

None of those points begins to found a suggestion that the Professor used AI to write his report

Deputy Chamber President Plimmer, Upper Tribunal Judge Blundell and Upper Tribunal Judge Landes, Upper Tribunal (Immigration and Asylum Chamber)

The report was still criticised, on other grounds

None of this means the expert had an easy hearing. The tribunal recorded that it was more concerned by other matters that emerged from cross-examination, including that he had not appended two of the operational documents he was relying on when describing gaps in identification, despite having produced six thousand pages of appendices, and that when asked what those documents said on a particular point he said he could guess, and then did so.

That is the distinction worth taking away. Criticism of an expert's method, sources and rigour is ordinary forensic work and the tribunal engaged with it. An allegation that the report was not written by the person who signed it is something else, and the panel treated it as a serious allegation requiring a proper foundation.

Why this reaches beyond one tribunal

Anyone running a construction or engineering dispute now works with expert reports drafted on machines that offer to write. Putting an authorship allegation in cross-examination, on nothing more than a repeated formula or a stilted sentence, is cheap and obvious. This decision sets out what happens when it is tried with no material behind it.

The reading for anyone instructing an expert is narrower. Repetition in a report often comes from repetition in the instructions, so the questions are worth drafting as carefully as the report that answers them. Missing the current edition of a source will cost weight whatever wrote the document. And a witness working in a second language will produce phrasing that reads oddly, which proves nothing about who typed it.

Three paragraphs, in a decision about something else entirely, and the first real guidance on the point.