Law360, London (May 29, 2026, 7:00 PM BST) -- A recent High Court ruling that exposed how lawyers had relied on fake artificial intelligence-generated legal authorities during insolvency proceedings has punctured an "arrogance" within the profession that AI hallucinations were a problem confined to smaller firms and inexperienced practitioners.

The incident, which involved a junior lawyer at Pinsent Masons LLP, one of the country's biggest law firms, exposed how easily confident-seeming but false material generated by artificial intelligence can reach the courts. The episode has resulted in the firm referring itself to the Solicitors Regulation Authority.

Lawyers say that the case of Anthony Cork and another v. Mark Smith serves as a warning that even elite firms are vulnerable to governance failures tied to generative AI.

The judgment adds to a growing body of global cases that have relied on nonexistent authorities generated by AI systems.

Sullivan & Cromwell told a New York bankruptcy court in April that a motion it had filed contained multiple AI hallucinations and errors. And, in London, Judge Victoria Sharp, president of the King's Bench Division, called on the profession to do more after two cases were blighted with dozens of fake case-law citations.

"There's been a bit of arrogance in the profession … thinking that this is only a problem for smaller firms or individual barristers at smaller sets," Ian De Freitas, a partner at Farrer & Co., said. "Both this case and the one in the States … illustrate that very large firms can have problems around this."

Reuben Vandercruyssen of Hogan Lovells said that this latest judgment from Judge Mark Mullen at the Insolvency and Companies Court stands out as "probably the clearest U.K. judgment so far on AI hallucinations in legal work, because it drills down into how the false material was put before the court.

"The more interesting point is how automation bias crept into the process," Vandercruyssen continued. "The output looked credible, and it fitted the answer the user was working towards."

Vandercruyssen said the problem was not simply that AI produced incorrect material — lawyers stopped critically interrogating answers that appeared polished and authoritative.

"A confident AI answer can cause the user to stop thinking critically," he said. "The real failure is not only that the AI got the law wrong. It is that human judgment was, in effect, handed over to the tool."

The High Court did not find the lawyers in contempt of court. Judge Mullen said instead that the referral to the SRA was an appropriate response. However, his judgment contained sharp criticism of the failures that led to fabricated authorities being relied upon in litigation.

Judge Mullen's judgment set out conversations between an associate identified as "LA" and the AI tool, which were submitted to the court by Pinsent Masons. The transcript, which ran to 59 pages, included a warning from Pinsents' internal AI tool that it was "not fully confident" with its answers and advised the lawyer to check its answers.

Judge Mullen wrote that he found the evidence "very troubling." He said the incident could have been avoided if the junior lawyer had checked any of the statutory provisions that the AI referred to.

"Legal professionals bear ultimate responsibility for their work and cannot outsource the process of legal research or of legal reasoning to an AI," the judgment reads. "It is a tool to be used with caution."

Pinsent Masons said in a statement that it had "apologized unreservedly to the court and are taking steps to strengthen our processes and oversight to ensure this does not happen again."

Lawyers say that the case highlights a growing concern within the profession that increasingly sophisticated AI systems can create a false sense of confidence, particularly when lawyers are under time pressure.

"This is happening increasingly commonly," De Freitas of Farrers said. "There's been a whole series of cases in this jurisdiction and others … where the court has been misled by a lawyer or lawyers not checking the output of AI properly."

Simon Walsh, a partner at SA Law, said the case exposed a practical weakness when law firms rely heavily on junior staff to use AI tools.

"A junior lawyer can ask an AI a question, but if they don't have the experience to check the answer … how would they know it's wrong?" Walsh said.

The case has also intensified debate over how firms supervise the use of generative AI tools internally. De Freitas said most major firms now have formal AI policies requiring lawyers to verify outputs before relying on them.

But Vandercruyssen said the judgment exposed a deeper issue in training and awareness, particularly among senior lawyers supervising AI-assisted work.

"The judgment says the senior associate on the case was at the time unaware of AI hallucinations,'" Vandercruyssen said. "That is striking, because it shows why AI training cannot just be aimed at trainees or junior lawyers.

"Everyone using, or supervising, the use of these tools needs to understand the basics: what they are good at, where they are dangerous, how hallucinations arise and why a polished answer is not the same as a correct answer," Vandercruyssen added. 

De Freitas also argued that firms should train not only their junior lawyers who use AI day-to-day. "It's also about training the partners and supervisors reviewing their work," De Freitas said.

Lawyers broadly agreed that the judgment should not be interpreted as an indictment of AI itself. They said that the case exposed weak workflows and poor governance over how AI systems are deployed in legal practice.

Courts and regulators have often distinguished between consumer AI products and approved enterprise tools. But the Pinsent Masons case demonstrated that even sanctioned tools could be inappropriate if used incorrectly.

Philip Young of Garfield.Law Ltd. — the first approved artificial intelligence law firm in England and Wales — took a narrower view of the significance of the ruling. The founder and chief executive of Garfield argued that the core issue was ordinary professional incompetence rather than systemic AI risk.

"I don't think the judgment is very interesting, and I feel sorry for the senior lawyers involved," Young said. "It's really a case about a junior lawyer being very sloppy."

Young noted that the AI system itself had reportedly warned the user to verify the material.

"The AI product told him or her to check the relevant material on more than one occasion and the junior lawyer did not do so," he said. "The only lesson … is that lawyers cannot be sloppy, but that has been true since the legal profession first evolved."

Others in the profession, however, argue that the frequency of AI-related incidents means that law firms can no longer treat hallucinations as isolated or unforeseeable mistakes. "It's definitely foreseeable now," De Freitas said. "Risk teams in firms, legal compliance officers … should all be aware of this now."

Vandercruyssen said that the underlying conditions that created the error were common throughout legal practice.

"Busy lawyers, heavy time pressure, credible-looking AI outputs, responsibility being pushed down to juniors, and work being treated as low risk because it looks procedural or unopposed — those conditions exist all the time," he said. "The reminder is that there are no truly low-risk contexts when material is being put before the court."

The Law Society has also used the judgment to reinforce warnings to firms over their use of generative AI in legal work.

Ian Jeffery, chief executive of the Law Society, acknowledged in a statement that AI could improve efficiency but said that "ultimate accountability for any legal submission remains with solicitors. We urge all firms to establish robust internal safeguards to ensure all outputs are thoroughly checked, and all staff are sufficiently trained."

The SRA is now expected to examine the case through the lens of existing professional obligations, including duties of competence, supervision and integrity.

The regulator might also face pressure to provide more concrete guidance to firms on acceptable governance practices for AI.

"The SRA has already faced criticism from the Law Society for not doing enough on AI," Vandercruyssen said. "This judgment gives it a very concrete example to work from."

--Editing by Ed Harris.

For a reprint of this article, please contact reprints@law360.com.