Three Tells, and Two of Them Pointed the Other Way.

In a UK asylum case, counsel for the Home Secretary suggested that an academic’s expert report was written by AI, on three grounds. The tribunal found no proper foundation. It then attached less weight to his reports, on separate grounds.

Three words in his own report were offered as a sign that a machine had written it: “to my opinion.” The report was expert evidence in a UK asylum appeal. In March 2026, at Field House, counsel for the Home Secretary put the point to its author, Professor Dimitris Dalakoglou, who writes in his second language. Giving oral evidence, he used the phrase again.

The case is LR (Male VOTs) Albania CG. The “CG” stands for country guidance: the panel was answering eleven listed issues about Albanian men who are victims of trafficking, not only one man’s appeal. Dalakoglou, the appellant’s expert, gave oral evidence through the whole of the fifth day. In the papers were 199 written questions from the Home Secretary, 69 more pages of answers, and a document he had written himself, entitled “Memo on the Nature and weight of the Expert Evidence.”

The tribunal reproduced how he introduced himself in his report. A British citizen. A Full Professor at Vrije Universiteit Amsterdam, “among the most frequently cited scholars in Balkan and Eastern European studies,” he wrote. A Seeger Fellowship at Princeton. Several years as a country-of-origin expert on Albania. It called that “a helpful summary.” Sixteen paragraphs later it called him “a respected authority on society, politics and migration in Albania.” In cross-examination, counsel put it to him that generative AI had produced the report. Counsel, the judgment records, “sought to develop at length” that point.

There were, the tribunal wrote, “essentially three bases upon which the allegation was founded”: phrases repeated within the report; his failure to mention the 2025 report in the US State Department’s annual trafficking series; and the phrase “to my opinion.” Call them tells. The third is three words long. None of the three concerned how the report had actually been made.

The list judges already had

The judiciary publishes a list of what AI-produced material looks like, in guidance addressed to judges and refreshed on October 31, 2025. The passage introducing it is about AI chatbots and unrepresented litigants. The guidance does not purport to govern expert evidence. The list names six indications: unfamiliar citations; parties citing different bodies of case law on the same issue; submissions that do not match the judge’s understanding of the law; American spelling or overseas cases; content that is “highly persuasive and well written” but contains “obvious substantive errors”; and an AI prompt left in by accident. Repeated phrasing, an omitted report and one expression are not on it.

The panel’s answer

The panel answered: “We do not consider there to have been any proper foundation for such a serious allegation.” On the second and third, it went further. They “actually suggest that the report was not written by AI, which would have been more likely to cite up-to-date material and use proper English.” He had overlooked the newer report, he explained, and, though a British citizen, “English is not his first language.” On the first ground, it noted “a degree of repetition” and recorded his explanation: overlap in the questions the appellant’s solicitors put to him. By the tribunal’s reasoning, counsel had offered as evidence of a machine two features a machine would have been less likely to show.

On the phrase, the tribunal said it would not say much: “we very much doubt that the question should even have been raised with an academic who was writing in his second language, and we note that it was an expression which he went on to use in his oral evidence.” So the supposed tell turned up again in the hearing. In the witness’s own mouth.

Counsel argued from three features of the report. The panel answered from the same three, but reversed two. The judgment’s account identifies no draft, file history or declaration of how the report was made. On that account, the more persuasive reading won, and remained a reading. The description is this publication’s, not the panel’s.

The other half of the judgment

The panel rejected more than the machine case. If the “pro-migrant advocacy” line meant that he lacked impartiality, the panel said, “there is no basis in the evidence for such a serious suggestion.” Even the Home Secretary’s case acknowledged both sides of the distinction: he was “clearly extensively published and cited but he had written nothing about human trafficking.” The tribunal’s assessment of his subject-specific evidence turned on the second half. “We were more concerned,” the panel wrote, “by certain other points which emerged clearly from cross-examination.” He had not appended two documents central to how Albania identifies victims, despite six thousand pages of appendices. Asked what they said about a victim’s consent, “The Professor stated that he could guess, and did so.” The panel formed “the clear impression that the Professor had scant knowledge of the improved NRM,” Albania’s National Referral Mechanism. The 2025 report he had missed, the omission offered as a sign of a machine, went to weight instead: it “will affect the weight which we give to the Professor’s report,” yet “does not begin to suggest that he used AI.”

His reports “rely on undisclosed interviews with un-named interviewees, which lessens the weight which we can properly attach to them.” Two other experts in the same hearing stood differently. The tribunal accepted that Dalakoglou “is not named anywhere in the 17,000 pages before us, apart from his own report,” unlike Ms Trimi and Dr Davy, who “feature prominently in background material which relates to the periods in which they were both heavily involved in anti-trafficking measures or scrutiny within Albania.” That difference, it added, “perhaps understandably reflects upon his broader-based expertise on the prevailing conditions in Albania.” The tribunal linked their prominence in the material to periods when both were heavily involved in anti-trafficking work. His evidence was, it concluded, “of limited assistance to us in relation to the prospective protection mechanisms in place upon return to Albania,” for those adult males accepted to be victims of trafficking.

The rule still being written

As of its June 2026 update, the Civil Justice Council’s working group had not finalised its position for the civil justice system in England and Wales. The update puts the direction of travel, “subject to finalisation,” toward “introducing proportionate transparency in relation to expert evidence.” This case was heard in the Upper Tribunal. On witness statements, the group asks whether existing mechanisms, “such as statements of truth, disclosure obligations and cross examination,” are sufficient. In this hearing cross-examination produced both results: the suggestion with no foundation and, in the tribunal’s words, the points that “emerged clearly.” Ireland’s High Court has adopted a different approach. Under Practice Direction HC142, in force since September 1, 2026, an expert report carries its own qualified declaration, and material use of generative AI must be disclosed (Edition 058). Under that rule, the authorship question has an address: the report itself.

The Agentics read

Agentics reads this pattern through Digital Derangement Syndrome™ (DDS™): AI-mediated systems failing to recognize legitimate authority. The characteristic that fits is Authority Misclassification, with its direction reversed. Earlier editions followed work that was not the author’s carrying the author’s name; this one follows an author’s work presumed not to be his.

Where the argument stops

The strain, plainly: no AI system acted on this professor. The judgment arose from human inference in cross-examination, so the fit to DDS™ rests on direction rather than mechanism. The tribunal held there was no proper foundation for the allegation; it did not find that he did not use AI. The allegation was counsel’s, not the panel’s. The appellant is anonymised by order; the only fact used about him is that his appeal was allowed on Refugee Convention grounds.

Nothing here says the judiciary’s guidance contradicts counsel. No statement from the professor, the Home Office, his university or counsel appears here. His report is not public. Whether permission to appeal has been sought was not established. The hearing is known here only through the judgment; whether drafts or file histories existed was not established. The mapping to Digital Derangement Syndrome™ is an editorial reading.