AI and disclosure: Too much information

Disclosure in large, complex commercial disputes routinely runs to hundreds of thousands, and often millions, of documents. Traditionally, these documents were reviewed manually by large legal teams, at significant cost and time, and with the potential for errors and inconsistencies to arise. 

In the early 2000s, a proliferation of email and electronic records meant that disclosure was becoming unmanageable. Old methods of manual review were too slow and increasingly unreliable. E-disclosure software involved keyword searches, which, while effective, were a relatively blunt instrument.

Technology assisted review (TAR) evolved out of e-disclosures. Instead of looking for keywords, examples of relevant and irrelevant documents were used to train the software to recognise patterns across wider datasets. Adoption was initially cautious, but by the late 2010s TAR had become part of the mainstream.

'We have had TAR for a long time, and it’s been encouraged by the courts and the practice direction, and this is just an extra layer of tech on top of that'

Reuben Vandercruyssen, Hogan Lovells

PD 57AD (the practice direction on disclosure) came into force on 1 October 2022 alongside the Disclosure Review Document (DRD). This is often the starting point for establishing a workflow for any disclosure exercise in the Business and Property Courts. The DRD form is in the appendix to the practice direction. It provides a structure for parties to a dispute to agree on how a disclosure will be handled. Not only does it record the disclosure methodology agreed between the parties, but it also creates accountability in the event of a later dispute.

The rapid evolution of AI in the legal sector has not left the disclosure process untouched. Generative AI review tools are able to offer a more efficient and cost-effective means of reviewing documents on complex disputes. While uptake is growing, until recently, practical guidance on how to use these tools effectively has lagged behind. 

The use of GenAI is no longer a speculative future tool, but something that is already being tested in review workflows, says Reuben Vandercruyssen, senior litigation associate at Hogan Lovells. ‘We have had TAR for a long time, and it’s been encouraged by the courts and the practice direction. This is just an extra layer of tech on top of that, which I think will change the baseline of what is expected when you are doing a big disclosure exercise.’

New guidance

The International Legal Technology Association (ILTA) published guidance for litigators in England and Wales in September 2025, indicating how GenAI should be approached in disclosure in the Business and Property Courts. This builds on the ILTA’s Active Learning in legal document processing.

Co-chaired by Fiona Campbell of Fieldfisher and Tom Whittaker of Burges Salmon, with contributions from several other authors, the Generative AI Best Practice Guide was initially an addendum to ILTA’s Active Learning Best Practice Guide. It is now a self-standing document, providing a procedural playbook on how GenAI can be used ‘responsibly in court-ordered disclosure exercises’. 

Campbell explains that PD 57AD does not address how emerging technologies should be used in a court-ordered disclosure. This lack of guidance has led to some litigation teams being hesitant to adopt AI, particularly given PD 57AD’s emphasis on transparency. ‘We felt that there was a need for a guide as to how GenAI should be used within guardrails,’ says Campbell. 

‘While PD 57AD defines TAR, it doesn’t address GenAI, because it was created before that even existed. While the guide references PD 57AD, its recommendations may be applied in full, or separately to suit different GenAI uses in legal practice,’ she adds. ‘It is drafted very much to be à la carte, pick and choose, speak to the other side about using GenAI without bothering the court, and these are the guardrails. 

‘It’s been drafted clearly enough for everyone to understand, and simple enough for someone who isn’t familiar with the technical aspects of eDiscovery to pick up, understand and agree.’

The ILTA guidance is already being used by firms in correspondence to justify approaches that are being taken, notes Vandercruyssen. ‘I think it will also get referred to in judgments this year. It is only six pages long and will develop over time, I imagine. It is quite conservative and “lawyerly” in its approach and talks about the risks.’ 

He hopes that users will also focus on ‘the upside, including the time savings… It will be interesting to see how it develops’. 

Human oversight

AI can work ‘brilliantly for first-level reviews on meaty complex disclosures’, says Campbell, although she emphasises that human oversight is still critical, even at this early stage. She references the decision Ayinde v London Borough of Haringey and Al-Haroun [June 2025], noting: ‘AI is not a decision-maker. Ayinde emphasised AI should not replace legal judgment.’ 

'I prefer to use GenAI for understanding a dataset if needed, creating a timeline, for summarising elements, so not necessarily for review'

Fiona Campbell, Fieldfisher

While GenAI will ‘have a greater sphere’ in terms of first-level review, it will still require oversight, checks and balances and bias tests, she adds.

However, its capabilities extend further than that, in her view. ‘I think to use it simply for review would be slightly misjudged,’ Campbell says. ‘Even if it is used for review, it is used over a pre-conditioned, pre-assessed dataset. It’s not used blindly. It’s used as an afterthought to the accepted methodologies, which are used day in and day out. It’s something that enhances rather than replaces. I prefer to use GenAI for understanding a dataset if needed, creating a timeline, for summarising elements, so not necessarily for review.’

Larger firms, such as Hogan Lovells, are currently using teams of paralegals to do the first level of review on a disclosure. However, Vandercruyssen predicts that in a few years this initial work will be carried out by AI, helping to speed up the process and potentially bring down costs. 

He acknowledges that teething issues will arise. There will be a need for tech teams across firms to collaborate, to ensure checks and balances are put in place. ‘The DRD will be very important for recording the detail of what you are doing,’ says Vandercruyssen. ‘I think expectations around the level of detail that needs to be included will evolve over time. For example, what prompts each team is using, and what quality control is being put in place.’ 

Market uptake

While there was previously a reluctance to use GenAI, lawyers note a marked change in attitude over the last 18 months. This has exceeded expectations.

Firms are using different eDiscovery platforms, with Everlaw, Relativity and Consilio among the most popular. In larger firms, GenAI features are bolted on to existing legal platforms that a firm is already running, rather than being deployed as standalone tools. 

Pushback on the use of GenAI tools is rare, notes Cambell – PD 57AD is built around early engagement, transparency and cooperation. Any pushback tends to be from smaller, or high street, law firms, she adds. That is ‘simply because they are coming from a different playing field and may possibly not have the level of understanding of these tools’. 

While smaller practices may be at a disadvantage due to not having the same in-house tech capabilities, their agility may in fact put them at an advantage in other respects, argues Vandercruyssen. 

'The distinction between us and some of the larger firms is that we are quite brave in using those tools, even when we are dealing with cases worth less than a £100,000'

Lewis Fairfax, Gibson & Co

‘I would suggest that quite a lot of them may be able to adopt AI tools more quickly and be more agile in how they use the technology,’ he says. ‘AI can be a great democratiser. My experience is that while the tools needed are not outrageously expensive, if you compare them to the cost of a disclosure exercise involving TAR, in some ways they may make it possible for smaller firms to run bigger disclosure exercises.’

This is a view shared by Lewis Fairfax, an associate at five-partner north-east firm Gibson & Co. ‘We are quite a small firm, so tend to lean quite heavily on disclosure platforms like Epiq and Consilio and have the same access as a larger law firm. On a multi-track, it’s pretty universal now to use one of those platforms,’ he says. ‘The distinction between us and some of the larger firms is that we are quite brave in using those tools, even when we are dealing with cases worth less than £100,000, which are in the fast-track or intermediate track, with less than 5,000 documents to review.

‘We are regularly opposite large London firms, and that’s never been an area of disparity. We might not have the manpower of those firms, but I think the days of large manual reviews are getting rarer and rarer. It can only level the playing field.’

Fairfax acknowledges that other firms of a similar size may not have had that ‘epiphany’ yet, even though they can improve efficiency even on smaller-value cases. 

‘The value of these AI tools is to identify relevant documents, which may not have been caught with keyword searches,’ Fairfax explains. While in the past, keywords would be applied to documents, if another pool of relevant documents came to light, the DRD would need to be amended by agreement between the parties. 

‘AI tools are really good at identifying relevant documents that are thematically related, building chronologies and helping a human reviewer to build a clearer picture as to whether what you’ve agreed in the DRD, for example, is what is needed to satisfy the disclosure obligations,’ he explains. 

Campbell reinforces this point: ‘In the context of a fraud investigation, where you are on a tight timeline, and you need base evidence to bring an injunction, AI can look holistically, it can chime in and draw a map of what’s going on. A manual review simply won’t be able to do that. And that’s I think the beauty of AI.’

The way forward

In November 2025, the Disclosure Review Working Group opened an online survey seeking views on how PD 57AD is working in practice. The deadline for responses was 3 February. 

Reuben notes that 20% of the survey questions related to the role of TAR/AI within disclosure, suggesting that these are now part of the mainstream operation of PD 57AD, and not a ‘niche’ legal tech conversation. 

‘What might come out of the working group is some sort of general statement that you should use technology to speed up and improve the quality of your review,’ he notes. ‘I think this idea of having to review a million documents is going to disappear. In my view, there are likely to be tweaks to the practice direction this year to indicate the kind of detail that needs to be provided.’ This will help to reduce ‘repeated negotiation friction – particularly around technology choices and what “good practice” transparency looks like’.

The Competition Appeal Tribunal’s disclosure ruling in Gormsen v Meta [2025] CAT 55 is also noteworthy, adds Vandercuyssen. Here, the tribunal took a ‘light-touch, case-management stance’. It declined to mandate any particular TAR tool or AI approach, but left the choice of methodology to the ‘designated solicitor’ with overall responsibility to the tribunal for the conduct of the disclosure exercise.

‘I suspect that this year there will be more judgments like Gormsen,’ Vandercruyssen says, ‘where you’ll get interesting pronouncements that guide the way, which will be along the lines of accountability, transparency and defensibility. There will be this focus on whoever is signing off on the disclosure being responsible for making sure everything is being done appropriately. Instead of mandating, they will be deferring to the lawyers.’

He adds: ‘Even before GenAI was used in a disclosure exercise, there has always been an accountable person, and that will remain the case.’ 

‘Ultimately it is just a tool,’ adds Campbell, referencing the deputy head of civil justice, Lord Justice Birss’s views on AI in civil justice and the latest Judicial Guidance on AI published in October. ‘It’s only a good tool if you already know the answer, and that is how I use it.’

Maria Shahid is a freelance journalist