Nineteenth century English engineer Isambard Kingdom Brunel understood that advances in technology bring new ways of thinking and said that rules should not be overly prescriptive, so as not to stifle innovation. He expressed this sentiment in 1948 in a letter to the Royal Commission on the Application of Iron in Railway Structures, warning against rules that might “embarrass and shackle the progress of improvement… by recording… the prejudices or errors of today.”
Practice Direction (PD) 57AD of the Civil Procedure Rules in England and Wales (the newer of the two regimes that govern disclosure, and the regime which is in operation in the Business and Property Courts) places technology front and centre by obligating that the parties’ legal representatives use it. However, it does not specify which technology should be used and, although not yet four years old, a lot has changed since PD 57AD came into force. Generative AI (Gen AI) is impacting elements of the legal industry, bringing with it both risks and opportunities.
Whilst PD 57AD provides that a party’s legal representative has a duty to promote the reliable, efficient, and cost-effective conduct of disclosure by using technology, Gen AI is not specifically mentioned in the current rules. The existing references to technology are purposefully non-exhaustive, as new technologies develop at pace, but the question arises as to whether the current framework is sufficiently flexible, or whether further rules and/or guidance from the courts are required.
Rerouting to Gen AI
If thought of as an umbrella term, AI has been used in disclosure exercises in this jurisdiction for years, but Gen AI is different.
The UK Courts and Tribunals Judiciary Guidance for Judicial Officer Holders regarding AI defines Gen AI as “A form of AI which generates new content, which can include text, images, sounds and computer code. Some generative AI tools are designed to take actions.”1 ILTA’s Generative AI Best Practice Guide (Outgoing Disclosure) highlights some of Gen AI’s features, including that GenAI models rely on non-deterministic machine learning models and require a prompt which will produce a response. Gen AI has a number of potential uses in the disclosure process, including for classification of documents (including based on relevancy), in the privilege review, and in the application of redactions.2
Since the launch of a public-facing Gen AI tool in November 2022, the eDiscovery industry has invested resources to harness this technology for use in the disclosure process.
Obstacles in the Path Ahead
When used responsibly, Gen AI may positively impact the disclosure process, not least in relation to potential cost and time savings.
However, under PD 57AD, the use of Gen AI poses some challenges:
- The necessity for cooperation: PD 57AD requires parties to cooperate with each other. The Civil Justice Council (CJC) for England and Wales has reported3 that anecdotal evidence suggests parties are currently cooperating in their use of AI in disclosure exercises, but the use of new technologies might lead to extended correspondence about their suitability and the agreed (or not) parameters around their use. Parties should consider putting safeguards in place so they are able to clearly articulate how Gen AI will be used. Parties should bear in mind the overarching objectives of reasonableness and proportionality. Debates about the use of AI where there is no reasonable basis for a concern that it will impact the integrity of the disclosure exercise may run counter to these objectives.
- The use of prompts: Historically, large disclosure exercises have been conducted using agreed keyword searches. The CJC notes that if parties instead choose to identify relevant material through the use of detailed prompts, questions may arise as to what level of disclosure is required in relation to the prompts themselves, including in relation to their testing and refinement. The circumstances in which privilege may attach to the wording of prompts inputted into Gen AI models used in litigation more broadly may also be debated and ultimately decided by the courts.
- Sufficient oversight: As is the case with the use of AI in any context, human oversight remains key. Where to draw that line in relation to large and complex disclosure exercises may be a point of contention.
Is a New Map Needed?
At the end of 2025, a working group established by the judiciary of England and Wales to review the operation of PD 57AD, sought feedback from the industry on the use of technology in disclosure.
A survey requested input on the operation of PD 57AD, including whether changes or modifications were needed to better adapt it to the continuing advances and increased use of technology-assisted review and AI in disclosure exercises. Requests were made for any suggested specific changes, along with whether a best practice guide would be helpful.
Additionally, the CJC consultation4 on the use of AI for preparing court documents closed recently. One of the topics covered was disclosure, and views were sought on whether there is a need to introduce a requirement that disclosure lists/statements have a section addressing the extent to which AI tools/software have been used. Whilst the CJC at present does not think there is a pressing need for this, it recognises that guidance may be required to deal with the use of AI in disclosure in due course, particularly in light of the associated reduction in the level of human oversight.
It therefore remains to be seen whether further rules and/or guidance may be introduced for cases being litigated before the courts of England and Wales. Given the ever-evolving nature of technology, prescriptive rules might become out of date quickly, but separate court-endorsed guidance may be helpful. However, perhaps Brunel was right, all those years ago, to question the futility of overly prescriptive rules.
3 Interim Report and Consultation – Use of AI for Preparing Court Documents
4 Interim Report and Consultation – Use of AI for Preparing Court Documents