AI and evidence in 2026, according to a former Lord Chief Justice, a Lloyds GC, and a leading arbitrator.
A panel at London International Disputes Week (LIDW) this week on AI, authenticity, and the future of evidence brought together Rt. Hon. The Lord Thomas of Cwmgiedd, former Lord Chief Justice and President of the Qatar International Court; Richard Blann, Head of Legal for Disputes and Operations at Lloyds Banking Group; Sophie Nappert, arbitrator at 3 Verulam Buildings; and our CEO Gregory Mostyn. Here is a summary of what they discussed.
How are courts responding to fabricated documents?
The instinct when confronted with deep fakes and AI-generated documents is to ask whether they should be admitted at all. In civil and arbitral proceedings, time spent arguing admissibility is generally time wasted. The productive question is what weight to give the material, which requires examining its reliability, provenance, and how it was generated.
Courts need to be prepared to compel full disclosure of how documents were created, including the underlying processes. Relying on proprietary software and declining to go further is not a position that should hold. The Post Office case shows what happens when it is assumed that computer-generated records are inherently reliable. The Criminal Procedure Rules Committee is already working on updated authentication requirements, and the Civil Procedure Rules Committee is expected to follow.
The design of any new authentication regime matters as much as the principle. If the burden falls on the receiving party to spot whether evidence has been fabricated, the cost of challenging authenticity rises, and smaller parties without the resources to run that exercise are the ones who suffer. The challenge is catching fabricated material without inflating the cost of every dispute in the process.
What is happening to disclosure obligations as AI makes document review cheaper?
Disclosure obligations have always been constrained by cost: you look as hard as the exercise justifies. AI removes that constraint. Because large datasets can be interrogated in seconds, courts will start requiring parties to look harder and further, on the basis that there is no longer a good reason not to. For organisations with large data estates, that is a significant change in what disclosure involves, and current proportionality arguments will not survive it.
Is privilege at risk when employees use AI tools?
Whether communications with AI tools attract legal professional privilege is unresolved. The US and UK are developing different approaches, and for cross-border disputes the variation is creating immediate uncertainty.
The problem runs deeper than a gap in the rules. Employees engage with AI tools without understanding they may be creating discoverable records. The tools are now embedded in voice assistants, phones, and Word before a document is even drafted. Legislative reform is unlikely to move fast enough to address this. Courts will have to do the work instead, across jurisdictions and without a consistent framework. Privilege was built around human communication. Whether it extends to interactions with automated tools is a question that requires fundamental review, not a patch.
Is common law or civil law better placed to handle AI-generated evidence?
Common law's flexibility gives courts the ability to adapt procedural rules quickly, without waiting for legislation. That is a genuine advantage when the challenge is moving faster than any legislature can respond to.
Civil law's more inquisitorial model offers something different: tribunals that take a proactive role, setting clear expectations about acceptable AI use and sanctioning departures from it, rather than waiting for parties to raise challenges. That approach could be particularly useful in arbitration, where the sanctions toolkit is narrower and active case management is the primary lever.
The more pressing question is whether any consistent international standard is achievable. Soft law has produced workable benchmarks in arbitration before, and a similar process focused on AI and evidence is the most realistic route to coherent guidance. The risk is that jurisdictions compete for claims by adopting permissive rules before that guidance arrives.
The question behind the questions
One theme connected every question: as AI takes over document analysis and fact-finding, the value of human judgment rises rather than falls. The premium shifts to strategy, advocacy, and the judgment about what the evidence actually means. The faster and more reliably the factual record can be established, the more of a lawyer's time is freed for the work that only a lawyer can do.