Barclays identified a number of outdated charges at the Land Registry that had been repaid but not formally discharged. Usually, these are manually removed by applying to the Land Registry, but to eliminate the time spent on this process, Barclays produced an AI system that would automatically issue discharges and apply to remove charges that were no longer required from registered titles. 

It took eleven months to develop the software and in September 2023, Barclays held a ‘go/no go’ meeting, in which they decided to roll out the system on approximately 41,000 charges which Barclays believed no longer secured live loans. This turned out to be wrong, and more than 5,000 live mortgages were accidentally discharged. This problem was only discovered when HM Land Registry and Barclays began receiving queries from borrowers who had been affected. 

Previous cases on discharge indicated that if a charge is discharged at the Land Registry, it ceases to exist in law and equity. This was a concern for Barclays as, clearly, a mistake had been made and this was not just a generous gift from Barclays to their borrowers. All parties, including the Terrys, accepted that a mistake had been made and there was likely an entitlement in principle to relief – no borrower/proprietor or third party presented a proper basis for denying the relief sought.

Most of the borrowers/proprietors and third-parties were not impacted by the mistake at all, but some found that the sale or remortgage of their property was delayed or negatively impacted. To assist those affected, Barclays gave voluntary undertakings to the court that if the mistake caused a loss, it would indemnify the borrower for those losses. 

Rather than blaming AI for the error, Barclays said that the mistake was made by the humans who made the decision to the remove the charges at the go/no go meeting. With this in mind, the court needed to be satisfied that the mistake was sufficiently grave to make it unconscionable for customers to benefit from it. This was a complicated test to satisfy without reviewing each case individually, as nearly 1,000 of the discharged charges raised issues of priority on the register. 

Due to the complicated nature of the situation, the Barclays case is said to be the first of its kind using the “bifurcated” procedure to dispose of a large number of related issues in a single representative claim. This is a process where the court could use a single claim to deal with common issues shared by all, and then use those decisions to deal with issues individual to specific parties. As a result, when necessary, any individuals from the represented class of persons (namely those represented by Terry & Terry) could be taken out into a separate hearing if their issues were different from those commonly held. 

In the instance of Terry and Terry, the High Court awarded summary judgment that it was unconscionable for them to benefit from the mistake and the court should therefore exercise its discretion to undo the discharge. This was on the basis that: 

• the AI programme was a lengthy and careful project;
• Barclays reacted promptly when they discovered what had happened; and 
• there was no sensible explanation for Barclays discharging charges which still had associated loans outstanding.

The same decision could not necessarily be given for the thousands of other discharged mortgages without reviewing the facts on a case-by-case basis. This was therefore pushed to future hearings.

It would have been interesting to see what view the courts would have come to had Barclays relied on a mistake in the AI system instead. If a mistake is presented as a software AI issue, or that the AI has generated and created its own rules, the court will be faced with a challenging issue, because it would not be capable of satisfying the current test for a qualifying mistake. In satisfying the test for a qualifying mistake, the requirement for ‘advertence’ presupposes that a human is making the error, not an unconscious machine or system. 

Until the law is extended to cover qualifying mistakes by those beyond human consciousness, this case is a warning that anyone looking for a legal remedy to undo mistakes involving AI would be well-advised to find a way to frame their claim as one of human error. 

As use of AI becomes more common, the law may have to adapt to reflect the challenges that AI presents, particularly if it reduces the scope for human error. Rapid adoption of AI may affect the fundamental principles of the law as it currently stands. This is particularly important for the banking and finance sector as there are thousands of clients and large companies who will want to use automatic programmes to make bulk transactions easier. In response to Terry & Terry, the industry is advised to do spot checks on the security in portfolios to ensure nothing has been inadvertently released.