In Cork v Smith the court found that a firm had misled it twice over, and the judge used the ruling to say something plain about where accountability sits when AI produces the work.
What the court found
The firm misled the court on two occasions. The first came when it put forward references that contained AI hallucinations, material that looked like authority but was not. The second compounded the first. Asked to explain, the firm used AI to produce an explanatory letter, and that letter was still wrong. An error made with a tool had been answered with the same tool and the same lack of checking, and the mistake carried straight through.
What the judge said
The judge found that a junior lawyer appeared to have almost entirely outsourced the thinking process to an AI program. That phrase does the work. The problem was not that a tool had been used, but that the judgement a lawyer is meant to bring had been handed over to the machine and never taken back. The supervising solicitor and the partner had failed to supervise properly, so the failure was not confined to the most junior person on the file. It ran up the chain.
The judge criticised a cavalier attitude to accuracy and said in terms that AI does not remove the need for proper research, thought and verification. The firm involved, Pinsent Masons, apologised and referred itself to the Solicitors Regulation Authority.
That the firm was a large and well resourced one is part of what the case teaches. This was not a hard pressed high street practice cutting corners for want of time. It was a firm with every means to check its work, and the checking still did not happen. The failure was one of habit and oversight rather than resource, which is why a firm of any size should read the ruling as being about them.
The principle to take from it
The ruling is a reminder rather than a new rule. The court expects a named person to remain accountable for what is filed, and the involvement of an AI tool is not a defence. A citation that a machine produced is your citation once it goes into a document with your name on it. The duty to the court sits on the person who signs, and it does not thin out because a program did the first draft.
The phrase about outsourcing the thinking process deserves to stay in mind, because it names the real risk in how firms use these tools. A model that drafts quickly and reads confidently invites the reader to accept it, and the more capable the tool becomes the stronger that invitation grows. The lawyer's job is to resist it, to treat the draft as a proposal to be tested rather than an answer to be adopted. When that discipline slips, the tool has not failed. The person using it has stopped doing the part only a person can do.
The detail worth holding onto is the second failure. Reaching for AI to explain an AI mistake, without a person checking the explanation against the source, turned one error into two. A firm should read this as a warning about process, not personalities. Put a person between the AI output and the court, make that person's check real, and keep a record that shows it happened. The accountability the court described is easier to carry when the firm has built the habit that supports it.
There is a point here for supervisors in particular. The junior in this case had the tool, the deadline and, it seems, too little oversight, which is the combination that produces this kind of failure. A supervisor who assumes that a capable young lawyer with a capable tool needs less watching has the risk exactly backwards. The easier a tool makes it to produce plausible work at speed, the more a firm needs someone senior to slow down at the point of sign-off and ask whether the work is right rather than whether it looks right.
The judiciary’s AI guidance sets out how the courts expect these tools to be handled.
No partner wants to explain a fabricated authority to a judge. If your checking habits have not caught up with your tools, we close that gap in a day: get in touch.