Tuyunuklu v. Akmal: UK Upper Tribunal Suspects AI Hallucination Behind False Reading of Rent Repayment Case | Advocate Prakhar

Case at a Glance

Background

Hakan Tuyunuklu, a tenant at a London flat, applied to the First-tier Tribunal (Property Chamber) for a Rent Repayment Order (RRO) against his landlord Mohammed Akmal, alleging harassment. The application covered the period from 10 April 2024. A complication arose: Tuyunuklu was in receipt of Universal Credit, and section 44 of the Housing and Planning Act 2016 limits a tenant’s ability to recover rent that was paid by Universal Credit — only the local housing authority can apply for an RRO in respect of that portion.

At the October 2025 FTT hearing, the tribunal asked Tuyunuklu to explain why he was entitled to an RRO given his Universal Credit status. Tuyunuklu argued that the Supreme Court’s decision in Rakusen v. Jepson [2023] UKSC 9 established that a tenant in receipt of Universal Credit can still apply for an RRO. When the panel indicated this was not what Rakusen said, Tuyunuklu accused them of bias, refused to remain at the hearing, and left. The FTT dismissed his application.

On appeal, the Upper Tribunal granted permission on the ground that only part of Tuyunuklu’s rent was paid by Universal Credit — he had made contributions from the non-housing elements of his Universal Credit as well. However, the appeal still failed on procedural grounds: he had never clearly communicated this key fact to the FTT, and his conduct at the hearing had made it impossible to continue.

The AI Issue

Upper Tribunal Judge Elizabeth Cooke’s treatment of the AI issue is measured but pointed. She noted that Rakusen v. Jepson does not support the proposition Tuyunuklu repeatedly advanced — that a tenant in receipt of Universal Credit can get back rent through an RRO. The Supreme Court decided Rakusen on a different point entirely. Judge Cooke then wrote: “From references to AI in the transcript I wonder if the appellant was misled by an AI hallucination.” This is judicial shorthand for: I think this litigant relied on AI-generated legal research that told him a case supported a position it does not support.

What the Court Decided

• FTT’s decision upholding dismissal not erroneous in law: the FTT was not aware that only part of the rent was paid by Universal Credit, because the appellant failed to communicate this clearly.
• Appellant’s conduct at the hearing — accusations of bias, demands for recusal, escalating verbal aggression — gave the panel sufficient reason to terminate the hearing.
• Appellant did not have a fair hearing in the colloquial sense, but this was entirely the result of his own conduct.
• Rakusen v. Jepson does not establish that a tenant in receipt of Universal Credit can apply for a full RRO — that proposition is “without foundation.”
• Appeal dismissed; FTT decision stands.

“From references to AI in the transcript I wonder if the appellant was misled by an AI hallucination. Whatever the reason, his point about Rakusen v. Jepsen is without foundation.”

— Upper Tribunal Judge Elizabeth Cooke, [2026] UKUT 174 (LC), 5 May 2026

The India Angle

Indian Law Equivalent

The closest Indian analogue to the UK’s Rent Repayment Order mechanism is found in the Model Tenancy Act, 2021, and state-level rent control legislation. Under Delhi’s Rent Control Act, 1958, or Maharashtra’s Maharashtra Rent Control Act, 1999, landlords who harass tenants or fail to maintain premises may face action before the Rent Controller and the relevant appellate authorities. Where tenants are receiving housing benefit or similar government assistance, the specific legal position on who can apply for repayment would need to be verified against the applicable state statute — AI-generated arguments on this point may not account for the specific limitations in state law. Similarly, under the Real Estate (Regulation and Development) Act, 2016 (RERA), litigants must be careful not to rely on AI summaries of RERA provisions that may not reflect state-specific implementing regulations.

Bar Council Rules

The Tuyunuklu case illustrates a problem that extends beyond professional responsibility to the welfare of the litigant. An unrepresented tenant who relied on AI-generated legal research that mischaracterised a Supreme Court decision lost a potentially meritorious case. BCI guidelines on legal aid and access to justice recognise the vulnerability of unrepresented litigants. The case makes an argument for advocates and legal aid bodies to provide basic AI-literacy education to self-represented litigants — explaining that AI can misrepresent even the highest court decisions.

Practical Advice for Indian Advocates

• When a client (or opponent) advances a legal argument based on a Supreme Court or High Court case, always verify the actual holding of the cited case — AI consistently misrepresents specific holdings, particularly in nuanced procedural or housing law contexts.
• Tenant clients using AI to research their housing rights should be advised to have their AI-generated legal research reviewed by a qualified advocate before acting on it — the cost of this verification is trivial compared to the cost of losing a valid claim on a procedural issue caused by AI-misunderstood law.
• In tribunal proceedings, courts and tribunals are becoming more alert to the “AI-misled litigant” phenomenon — when you observe an opponent advancing a clearly wrong reading of a real case, consider whether a polite factual correction in writing might resolve the point without escalation.

Quick Takeaways

• AI can misrepresent the holding of even Supreme Court decisions — the hallucination is not always an invented case, but sometimes a real case misread.
• A litigant who relies on AI’s mischaracterisation of a case, then becomes aggressive when the court corrects them, undermines their own case and may cause a hearing to be terminated.
• UK courts are now noting AI involvement from transcript evidence — the judicial record of AI-related issues is becoming a new dimension of appeal scrutiny.

Deep Dive: When AI Misreads Real Cases — The Subtle Hallucination

The Tuyunuklu case illustrates a type of AI error that is subtler and in some ways more dangerous than generating a completely fictitious case citation. Here, the case (Rakusen v. Jepson) was real, the citation was correct, and the subject matter was at least tangentially relevant — the case does involve rent repayment orders and Universal Credit. What AI hallucinated was the holding: it told Tuyunuklu (or implied through its summary) that the case authorised his application, when the Supreme Court had decided Rakusen on a different point and explicitly noted that a tenant whose rent is paid by Universal Credit cannot recover it.

This type of holding-misrepresentation is more likely to fool even a moderately attentive reader than a fully invented citation. An invented case can be caught in seconds. A real case that is summarised incorrectly requires reading the case — which is precisely what many AI users skip. The Tuyunuklu transcript apparently contained enough references to AI-generated research that Judge Cooke could identify it as a likely source of the misconception. But the typical adversary or tribunal would not necessarily catch a real case being cited for the wrong proposition unless they read the full decision.

The practical defence is not simply “check that the case exists” — it is “read the relevant parts of the case and verify that the cited proposition appears in the judgment.” This is a more demanding standard, but it is the only one that catches holding-misrepresentations. For advocates who use AI in practice, the workflow should be: (1) AI identifies potentially relevant cases; (2) advocate locates each case in an authenticated database; (3) advocate reads the headnotes and relevant passages; (4) advocate confirms the specific proposition they are citing is accurately stated. Steps 2-4 cannot be performed by AI — they require human judgment applied to primary sources.