In this article, Chris Felton and Katie Dyson of Gardner Leader discuss the application of generative AI (GenAI) tools to the regime under Practice Direction (PD) 57AD (Disclosure in the Business and Property Courts) and set out key considerations for dispute resolution practitioners.
Paragraph 3.2(3) of Practice Direction (PD) 57AD (Disclosure in the Business and Property Courts) requires litigators to cooperate to ensure that disclosure is “reliable, efficient and cost-effective”, including through the use of technology. Generative AI (GenAI) has already shown significant promise in this regard. Predictive coding tools such as Technology Assisted Review and Continuous Active Learning are now routinely deployed in largescale litigation and there is growing pressure from both clients and the courts to justify any decision not to use them.
As GenAI continues to advance, it offers not only improved tools for practitioners, but also an opportunity to reshape an often adversarial and costly process and, in doing so, support the collaborative model envisaged by PD 57AD.
In the Business and Property Courts, parties are expected to engage with one another on disclosure from an early stage. Where extended disclosure is sought, a Disclosure Review Document must be prepared, setting out an agreed List of Issues for Disclosure and the composition of each party’s dataset. Where relevant, parties must also propose proportionate limitations on the scope of their review, including through the use of targeted keyword search terms.
Effective search parameters are integral to any disclosure exercise involving substantial volumes of documents; yet their formulation is often laboured and contentious, and the results are imperfect. This is especially so where the dispute spans several years, raises multiple or wide-ranging issues or involves allegations of dishonesty. In such cases, parties must balance the need to cast a sufficiently wide net against the risk of inflating costs through the capture of an excessive number of false positives. Achieving that balance typically involves a prolonged and expensive process of data sampling, iterative testing and negotiation.
The outcome is often the reduction of complex issues into rigid Boolean search strings, combined with broader metadata-driven parameters such as date ranges. These can prove unreliable, particularly where documents are no longer in their native format. The result is an increased need for manual intervention, higher costs and a greater risk that key material will either be missed entirely or buried within an overly broad review population.
GenAI-powered tools are beginning to offer more flexible solutions to this. Rather than relying on rigid keyword search terms, they can interpret natural language prompts to analyse large volumes of data. Those prompts can be tethered to the Issues for Disclosure, allowing the AI to assess documents in context (based on their meaning and substance) rather than through metadata or surface-level keywords alone.
The implications are significant. While traditional keyword-based methods may miss relevant material where no obvious search terms exist, AI can identify patterns of language or emotional tone that point to relevance without relying on specific words or phrases. This allows for documents to be identified based on sentiment or tone which is a potentially critical capability in cases involving allegations of fraud, dishonesty, collusion or misconduct.
The same benefits apply to foreign language documents. Traditional review methods may overlook relevant material unless suitable search terms are agreed in the original language and even then, translation can introduce cost, delay and complexity. These difficulties also arise when formulating search terms: without a sufficient grounding in the language, the iterative testing and data sampling needed to develop keyword search terms that are both wide-ranging and proportionate becomes difficult. Thematic tools powered by GenAI are increasingly not constrained in this way: they can identify documents expressing similar ideas or sentiments, regardless of the language in which they appear.
We must also be alert to the risks posed by GenAI and the potential consequences of failing to engage with it meaningfully. Deepfakes (highly realistic synthetic or manipulated documents) may be presented to parties, their lawyers or the court as genuine, with a view to influencing the outcome of a case. In this context, GenAI may be both part of the problem and part of the solution. Emerging tools are increasingly capable of identifying anomalies that are not otherwise apparent and it is reasonable to assume that such functionality may soon be integrated into disclosure review platforms. These tools may help flag materials whose content, tone or provenance warrants closer scrutiny.
Such advances will inevitably reshape what practitioners, clients and the courts consider to be a reasonable and defensible disclosure strategy. As these tools become more widespread, and their capabilities better understood, it will become increasingly difficult to justify a departure from them. Just as the courts now expect parties to explain any preference for manual review over accepted technology-assisted methods, similar expectations may soon apply to GenAI.
This also aligns with the issue-based approach to disclosure favoured by the courts. Disclosure remains a resource-intensive but critical part of the litigation process: it ensures that the courts have the material they need to fairly determine the claim. Responsibility for a robust and compliant disclosure exercise cannot be delegated; lawyers will always have a critical role in managing that process. However, time and cost should be directed towards matters that genuinely assist the court’s resolution of the dispute. Satellite debates (particularly those that could be avoided through the appropriate use of technology) should be minimised.
In largescale disclosure exercises, many law firms already rely on sophisticated review platforms and specialists with expertise in search optimisation. That is unlikely to change. What must change, and is already changing, is our collective understanding of the potential of emerging tools like GenAI. Both practitioners and the courts need a clear grasp of what these systems can and cannot do in order to engage meaningfully on questions of scope, proportionality and cost.
The pace of development over the past few years has been striking and there is no reason to think it will slow. These tools have the potential to enhance access to justice, improve efficiency and support fairer, faster outcomes. We have a duty as litigators and as officers of the court to engage seriously with the opportunities they present.
Reproduced from Practical Law with the permission of the publishers. For further information visit www.practicallaw.com