AI Regulation Tracker  /  Court ruling

UK court warns the CPS after it cited two non-existent authorities, attributed to AI, in an extradition appeal

In Tobosaru and Tofan v Romania, decided on July 8, 2026, the England and Wales High Court found that the Crown Prosecution Service had cited two authorities that do not exist. The CPS attributed the error to AI. The court issued a warning and declined to impose any sanction.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 13, 2026  ·  Last updated August 4, 2026

A court that finds fabricated citations in front of it has a range of responses available, from a plain warning through to referral or a finding of contempt. Here the England and Wales High Court chose the mildest end of that range. In Tobosaru and Tofan v Romania, an extradition appeal decided on July 8, 2026, the court found that the CPS submissions contained two authorities that do not exist. The CPS acknowledged the citations, attributed the error to AI, and accepted that the reviewing lawyer had not verified the references before filing. The court accepted the apology, recorded the remedial steps, and set the episode out as a warning rather than a punishment.

What the court actually found

The court found that two case citations relied on in the submissions did not correspond to real judgments. The false citations were identified before the appeal hearing and removed, so they did not affect the argument or the court's decision, and both appeals were dismissed on their merits. The CPS acknowledged citing the two non-existent authorities, accepted that they had likely come from a generative AI tool, and told the court that the operative cause was the human failure to verify: the lawyer reviewing the submission had not checked their accuracy before it was filed. The court agreed with that characterisation. It did not treat generative AI as an excuse. On the record before it, the court accepted the apology given for the CPS and the assurance that no one had tried to mislead it.

The judgment then delivered its warning in measured terms. As the court put it: "It would be naive to assume that there will not be an increasing use of artificial intelligence in legal work in future." The same passage adds that such use may be necessary and beneficial, and that the episode "highlights the risks of its use without appropriate oversight particularly for legal research." That framing matters. The court is not against AI in legal work. It is against filing what a tool produces without checking it.

A warning, not a sanction

This is the part to be precise about. The court declined to impose any sanction. There was no finding of contempt, no referral, and no penalty. The judgment does not name or fault any individual lawyer as personally responsible, and nothing in it should be read as discipline against a named person. The court's stated reason for setting the matter out at all was the serious consequences an error of this kind could have had in other circumstances, where a fabricated authority might slip through and shape an outcome. The signal is aimed at the profession, not at one advocate.

What the CPS did about it

The remedial steps are the part of this judgment that other organisations can copy. The CPS apologised to the court. It then carried out a full internal review, and had senior managers examine 78 other cases handled by the same lawyer. No similar problems were found in any of them, and the lawyer's work was otherwise assessed as being of a good standard. That is the shape of a proportionate response to a fabricated citation: apologise, correct the record, and go back through the rest of the file to establish whether the failure was a one-off or a pattern.

The operational lesson follows directly. Legal teams need a documented verification step between AI-assisted drafting and any filing. Every cited authority should be traced to the actual judgment on an authoritative database before it enters a pleading, skeleton, or submission, and supervisors need a checking routine for junior work that assumes citations may be fabricated.

Why a US professional should track this

The rule underneath this is not uniquely British. Every US attorney carries a non-delegable duty to verify the authorities they put before a court, and that duty does not transfer to a model. A generative tool can invent a citation that looks real, complete with a plausible reporter reference, and the only thing standing between that invention and a filing is a human check. US courts have already sanctioned lawyers for exactly this, so the English warning reads less as foreign news and more as the same lesson from another bench.

The practical takeaway is a workflow one. If your practice uses AI for research or drafting, the verification step is not optional and it cannot be delegated to the tool that produced the text. Confirm that every authority exists, says what the draft claims, and remains good law, before anything is filed. The court here was lenient. The next bench, in the next case, may not be.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any ruling applies to your situation with qualified counsel in the relevant jurisdiction.