Part 3 - AI for legal professionals: Document review and disclosure
One of the most significant areas where AI is already making a tangible impact, in the context of the legal profession, is in document review and disclosure. As lawyers face ever-increasing volumes of data, AI tools are offering faster, more cost-effective, and potentially more accurate alternatives to traditional manual review processes. The third instalment in our series on AI for legal professionals, this article focuses on the current role of AI in the disclosure process and what the future might hold for lawyers willing to embrace new technology.
The role of AI in disclosure
AI is not a futuristic concept in disclosure. Technology Assisted Review (“TAR”) involves the use of supervised machine learning, where a skilled lawyer trains an AI system on a set of seed data to recognise responsive and unresponsive documents. The AI system then applies what it has learned to the larger document population, classifying documents by their likely relevance or responsiveness.
This has been available to lawyers for several years now. Predictive coding, a species of TAR which has become increasingly prevalent in commercial disputes, was given the judicial green light in 2016 in the case of Pyrrho Investments v MWB Property [2016] EWHC 256 (Ch) and was further endorsed in subsequent cases which note its potential to reduce costs and increase efficiency. It allows the AI system to learn from human decisions to predict which documents in a dataset are likely to be relevant.
The value of TAR, and predictive coding, is clear. It goes well beyond simple keyword searches, identifying patterns and concepts relevant to the case in question. The more input that is provided, the more the AI system learns and the better decisions it makes, which can significantly expedite review processes involving vast volumes of information. As lawyers code more documents, the AI system becomes increasingly precise, enabling the faster identification of key material while reducing the burden of irrelevant data.
As well as predictive coding, document review platforms now commonly include features such as email threading and de-duplication, which help reduce the volume of material that must be manually reviewed by removing duplicates and grouping related documents.
Other, more sophisticated platforms which integrate generative AI are being developed and adopted, which allow lawyers to search for documents by asking natural language questions rather than training models using manually reviewed documents. Such tools allow for the retrieval of documents that may be relevant even if they do not contain pre-defined terms or deviate from established patterns.
Generative AI-enabled tools differ in that they focus on replicating patterns in phraseology, grammar and syntax, predicting the next word in a sentence based on the preceding context. Generative AI’s ability to create new content in this way also enables it to produce summaries of documents, suggest document classifications and generate review notes to assist reviewers. Sentiment analysis and anomaly detection can also be used in some cases to identify documents containing emotionally charged language which can assist lawyers identify potential "hot documents" more quickly.
What the future holds
While TAR tools generally rely on traditional machine learning, generative AI systems, trained on vast datasets, can go further. Whatever functions AI tools can perform now, they will be able to deliver them more effectively as the technology improves.
Looking ahead, the integration of generative AI into document review platforms will likely combine all stages of the disclosure process, from document retrieval to review to production, within a single AI-enabled environment. This more streamlined process will likely incorporate “agentic” features such as AI co-pilots, interactive assistants embedded within review platforms which offer dynamic guidance during the review process. These tools might suggest the next best documents to review, explain the reasoning behind certain predictions or help reviewers remain consistent in their coding decisions. Such tools will help to support junior lawyers, while allowing senior lawyers to focus on high-level strategic issues.
As the use of AI in document review and disclosure becomes more widespread, it is also likely that judicial expectations will shift. Whereas AI is currently an optional feature parties might consider adopting to save time and money, courts may begin to expect the parties’ use of AI tools in complex or data-heavy cases, particularly where the proportionality of the review process is under scrutiny. The court already encourages cooperation between parties and their representatives for efficiency in disclosure planning (Practice Direction 57AD). It is not difficult to imagine future revisions to the Civil Procedure Rules such that parties are expected to disclose whether they are using TAR or other AI tools as part of their review, and an explanation on how these tools are being deployed.
The rise of AI will also bring new procedural and evidential challenges. Today, parties often argue over search terms. Sometimes these arguments even result in satellite litigation. As generative and agentic AI tools are increasingly used to identify, summarise and classify documents, the crafting of the prompts that guide those tools will be crucial. The output of AI systems can vary significantly depending on the way the question is framed. Inconsistent or biased prompting could affect the results of the disclosure exercise, just as incomplete or overly narrow search terms might today. Parties may begin to challenge not only the outcome of an AI-led review but also the manner in which the AI system was instructed. This could lead to disputes over the suitability, transparency and reproducibility of prompts.
Over time, this may lead to the issuance of judicial guidance or even attempts to standardise the way parties approach AI-led disclosure. It is conceivable that parties will be expected to agree a set of core prompts in advance or to disclose example prompts used in the review process. Where disputes arise, the court may be asked to adjudicate on the adequacy of AI inputs in much the same way it currently rules on the reasonableness of search strategies.
Practical considerations
However, a growing reliance on AI does raise practical and ethical concerns. First, AI systems are only as good as the data they are trained on and the parameters they are provided with. If the training data or the seed data reviewed by a lawyer is skewed or not representative in some way, the resulting predictions may be unreliable. Lawyers must therefore be mindful of bias and errors.
Second, no matter how sophisticated an AI tool is, it is just a tool. AI tools are not a substitute for legal judgment, nor are they poised to displace lawyers. The use of AI may help lawyers to identify relevant documents or summarise their content but AI cannot reliably determine whether a document is privileged, for example. Ultimately, lawyers are responsible for the conduct of a disclosure exercise, and the court – and the client – will expect qualified legal professionals to exercise oversight and apply their legal judgment.
AI-enabled review platforms are increasingly equipped with source identifiers for this reason, allowing lawyers to see which documents or passages the AI system has relied on to reach its conclusions. These citation features will be particularly useful if an AI tool is used to assist with the drafting of the Disclosure Review Document, for example – as will they necessarily become an increasingly important feature of AI tools if there is to be trust in AI-led disclosure processes.
Comment
AI’s role in the disclosure process will deepen as the use of generative AI-enabled review platforms and agentic AI tools increases. These developments will not replace lawyers’ roles in the process but they will assist lawyers, allowing them to focus less on volume and more on strategy. That said, AI’s role in the process should not be taken for granted and lawyers’ judgment will remain crucial to ensure that the use of AI makes the process more efficient rather than undermining it.
In the next edition in our series on AI for legal professionals, we will explore how AI can be used by lawyers, litigation funders and mediators to inform their litigation strategies.
Footnotes
[1] See our previous instalments
25 September 2025 Part 1 - AI for legal professionals: Where to start? |
09 October 2025 Part 2 - AI for legal professionals: Hallucinations |
About Hausfeld’s Digital Markets expertise
The Hausfeld Digital Markets Hub