High Court expects ‘human supervision’ of GenAI
New guidance on the use of generative AI (GenAI) tools in the High Court states that all court users are responsible for the content of their court documents.
A practice direction from the President of the High Court Mr Justice David Barniville says that, while GenAI tools can assist in the efficient conduct of litigation, their use must be “appropriate and undertaken responsibly and with due care”.
It warns that such use should not “adversely affect the integrity of court processes and the proper administration of justice”.
It adds that the use of GenAI does not diminish or alter any obligation owed to the court by parties, legal practitioners, witnesses, experts, or any other person involved in proceedings.
‘Appropriately verified’
Practice direction HC 142, which comes into effect on 1 September, says that any person responsible for a court document must ensure that any material generated with the help of AI has been “appropriately and independently verified” before it is used in proceedings.
“Any limitation on the ability to do so must be disclosed to the court,” it adds.
The guidance sets out how content produced using GenAI should be verified, stating that this must be done “under human control and supervision”.
For legal practitioners, the direction says that they must exercise oversight and verify the accuracy and suitability of information provided by any GenAI system, adding that they must also ensure that their clients are informed of and understand their duty to the court.
The guidance sets out the court’s expectations on the appropriate and responsible use of GenAI tools in legal proceedings and the implications of the technology for practice and procedure.
Bias and inaccuracy among risks
The practice direction highlights the risks associated with the use of GenAI tools – including the risk of inaccuracy – and discusses the duties and obligations of parties and their legal representatives when these tools are used.
The risks listed include:
- Inaccuracy – including the risk that AI will generate fabricated or manipulated material, or fictitious citations (known as ‘hallucinations’),
- Bias and incompleteness, arising from the data on which the tool has been trained and how it has been processed,
- Factual errors,
- Confidentiality and privacy risks, and
- Legal privilege, with the guidance warning that information entered into GenAI tools or systems may be required to be disclosed in legal proceedings.
The guidance applies to all documents used in civil proceedings and all persons involved in court proceedings.
It also applies to all proceedings before the court “whether instituted before or after its commencement” and “all court documents prepared, produced, or generated after the date of commencement”.