AI IN LEGAL DISCOVERY

Ethics, Fairness, and the New Inequality of Arms

How AI is tilting the balance in legal discovery…

Corporate litigation has, in effect, become a software licensing contest. If you’re a CEO reading this, take some comfort – your legal strategy may now depend rather less on the righteousness of your case than on whether your legal team’s tools cost more than your opponent’s annual stationery budget. For years, the profession clung to the reassuring fiction that justice was blind, the courts level and “equality of arms” something more than an attractive constitutional phrase. Then the data multiplied. In an age in which a single executive can generate 10,000 emails, WhatsApp messages and Slack threads before lunch, manual review is no longer heroic. It’s merely untenable. Enter electronic discovery and its gleaming new engine – AI.

That leaves the modern C-suite with an awkward question – does justice belong to the side with the best algorithm? The issue is no longer whether lawyers have read the documents. It’s whether an AI system has sorted, ranked, or sidelined them in ways that are efficient, defensible and, one hopes, not catastrophically convenient. When billable hours yield to processing power, justice starts to look less like a principle and more like a line item on a technology invoice.

The old image of the sleep-deprived junior associate riffling through dusty boxes has not so much faded as been summarily replaced. The battleground is now digital, largely invisible, and heavily automated. For senior executives, who demand a more sober view of litigation risk. Legal readiness can no longer be measured by the pedigree of external counsel alone. If your opponent is deploying sophisticated machine-learning tools while your internal teams are still trying to export PST files without mangling the metadata, you’re not entering a contest on equal terms.

• Psssst – A PST file (Personal Storage Table) is a proprietary file format used by Microsoft Outlook to store local copies of emails, calendar events, contacts, and tasks. It allows users to archive or back up mailbox data to their computer’s hard drive, freeing up server space.

More importantly, the shift goes to the heart of legal ethics. If a court is asked to determine the truth on the basis of evidence that no human being has meaningfully reviewed from end to end, tough questions follow. Traditional cross-examination is of limited use against a model. One can’t ask a classifier why it treated a critical chain of executive WhatsApp messages as irrelevant, nor can one meaningfully test the proprietary logic behind a vendor’s claims of accuracy. The result is an uncomfortable possibility – that legal reality may be shaped, at least in part, by computational advantage rather than the merits alone.

The Global Landscape – Algorithmic Goliaths and the Illusion of Fairness

Courts in a number of jurisdictions have spent more than a decade warming to technology-assisted review and predictive coding. In the United States, Da Silva Moore v Publicis Groupe is widely cited as an early judicial endorsement of computer-assisted review. The attraction was practical rather than mystical – human review of millions of digital records is slow, costly, and often inconsistent. By that stage, the idea that armies of lawyers should read every electronic artefact by hand already looked less principled than nostalgic.

England followed with Pyrrho Investments Ltd v MWB Property Ltd, a decision often cited for accepting predictive coding as a proportionate means of handling large volumes of material. The promise was appealing – lower costs, greater consistency, and a more manageable route through sprawling corporate data. Inevitably, the technology was presented as a great leveller. Corporate counsel were invited to believe that software would democratise disclosure while trimming the less charming excesses of the billable hour.

In practice, however, the technology has also intensified an arms race. In major commercial disputes, well-funded parties can deploy advanced review platforms that go far beyond crude keyword matching. These systems identify patterns, cluster concepts, and prioritise potentially relevant material at a scale no human team can match. That may improve efficiency, but they may also widen the gap between those who can afford such tools and those who can’t. Increase the divide between the haves and the have nots.

Less well-resourced litigants – by contrast – may still rely on linear review and basic keyword searches. The imbalance is obvious. When one party has a precision instrument and the other has a glorified filing torch, the outcome of discovery may be shaped long before anyone reaches a courtroom. At that point, the “equality of arms” begins to look less like a doctrine and more like a fond historical reference.

The Digital Divide – A Very South African Tech Dilemma

If the global story resembles a high-end legal thriller, the South African version adds the sort of practical constraint that tends to ruin everyone’s cinematic posture. Here the digital divide is not an abstract concern. It is an everyday operational fact, and it runs straight through the legal system.

South Africa presents a striking paradox. On one hand, the country has a sophisticated corporate sector and highly capable legal practices, supported by modern matter-management and litigation-support tools. On the other, the formal rules governing discovery have been slower to adapt to the realities of electronically stored information. That’s not merely a theoretical complaint. South African legal commentary has, for years, pointed to the strain between conventional discovery rules and modern digital records, particularly where metadata, deleted material and native-format documents are concerned.

The irony is easiest to see in an ordinary commercial dispute in Gauteng or the Western Cape. A multinational company may be able to map a case across vast volumes of data in hours, running automated searches, deduplication, and privilege workflows with impressive ease. Across the aisle, a smaller opponent may still be contending with unreliable connectivity, load shedding and the sort of manual document handling that turns litigation into a test of both patience. And toner.

• Pssst – Deduplication (often called “dedupe”) is the process of identifying and eliminating redundant copies of identical information within a data system, while maintaining system integrity. Instead of saving duplicate files or data blocks, it stores a single instance and replaces the extras with references that point to the original.

That is not equality of arms. It is a structural imbalance in which the better resourced litigant can overwhelm the other by volume alone. Existing procedural tools may not always equip courts to insist on technological parity, or anything close to it. The risk is that data dumping becomes a perfectly respectable tactic – not elegant, certainly, but effective in the way many unpleasant tactics are.

The Black Box and the Case for Mandating AI Transparency

This leads to the harder ethical question. When a party relies on a proprietary AI tool to discharge discovery obligations, the opposing side is asked to place considerable trust in a process it can’t fully inspect. The producing party says, in effect, that the system has reviewed millions of documents, identified what matters, and excluded what does not. That may be true. It may also be difficult to test properly from the outside.

The problem with any black box is not necessarily that it’s wrong, but that its errors can be hard to detect and its assumptions harder to challenge. A model trained on a narrow or unrepresentative set of examples may reproduce those biases with great confidence and admirable speed. In law, that’s not a virtue. It’s merely a more efficient route to the wrong answer.

That concern has pushed many legal commentators, and some courts, towards greater emphasis on cooperation, disclosure protocols, and validation. The underlying point is simple enough – if parties are going to rely on technology-assisted review, they should be able to explain the process in a way that’s intelligible, proportionate, and open to challenge. The machinery needn’t be surrendered in full, but it also can’t be treated as sacred and beyond question.

For South African practitioners, the lesson is straightforward. If procedural reform continues to move towards fuller recognition of electronic discovery, the debate can’t stop at defining an electronic document. It must also confront transparency, validation, and fairness. If a bank, insurer, or mining house uses AI tools to sift through records in a major dispute, there must be some principled basis on which the process can be explained, tested and, where necessary, challenged. Without that, the Constitution’s section 34 promise of access to courts risks sounding impressive while doing rather less work than advertised.

South African courts have, meanwhile, shown little appetite for indulgence where lawyers use AI carelessly. Commentary in De Rebus discusses cases including Mavundla v MEC: Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal and Others as part of a broader warning against unverified AI-generated authorities and submissions. The principle is plain enough – practitioners remain responsible for what they place before the court, regardless of whether the draft began with a keyboard or a prompt.

The point extends beyond fabricated citations. It suggests a broader judicial insistence on accountability in the digital age. A corporate litigant can’t sensibly respond to a discovery failure by explaining that the software was “having a moment”. The burden remains human. If a legal team uses automated tools, it’s still answerable for the quality and integrity of the result. Technology may accelerate judgment.  It does not replace it.

Transition to Practice – When the Gavel Falls

All of which is intellectually stimulating over dinner and perfectly suited to conference panels. For executives and general counsel, however, the issue is practical. What happens when the debate leaves the seminar room, the judge takes the bench, and pleasant abstractions give way to deadlines, disclosure failures, and actual risk?

Once a commercial dispute reaches trial, the digital divide stops being theoretical and becomes operational. Courts aren’t equipped, nor inclined, to referee an endless technical quarrel between competing software vendors. Judges want evidence that is intelligible, properly authenticated, and usable. They’re unlikely to be charmed by the submission that your review process was almost state of the art, apart from the small catastrophe in the metadata.

In our next instalment, we move from principle to practice. We will examine how corporate legal teams can make the passage from AI-assisted discovery to courtroom proof, reduce the risk of being outmatched by a better organised opponent and ensure that, when the matter is finally argued, technology serves the case rather than quietly becoming it.

AJS’s Version 6 already includes some AI compatibility, which we regard less as a grand reveal than as a sensible starting point. The more interesting work lies in developing ways for legal teams to use AI that are practical, well governed, and genuinely useful. The ambition isn’t to replace professional judgement, still less to imitate it, but to support it where technology can do so responsibly.

For legal teams thinking seriously about modernising their operating model, adopting legal technology more intelligently or extracting more value from systems already in place, these questions are no longer theoretical. They are increasingly part of the job. AJS works with organisations facing precisely that transition, with an emphasis on tools that strengthen professional judgement and expose where legacy workflows have outlived their usefulness.

AJS helps legal teams modernise with a clear view of what technology should do, and what it should not pretend to do.

– Written by Alicia Koch on behalf of AJS

(Sources used, and to whom we owe thanks: Global Citizen; De Rebus on the ethical use of AI in legal practice; De Rebus on discovery of electronic documents and attorneys’ obligations; Harrisons eDiscovery Consulting; Fortinet and Crowell Data Law)