AI and the Integrity of Disclosure: Are the Existing Procedural Safeguards Fit for Purpose?

The International Bar Association recently published an article written by David Hopkins (2013) and Santosh Carvalho (2023) on the challenges AI may pose to the disclosure of documents and other evidence in common law adversarial litigation. They set out a synopsis of their view here, drafted with the assistance of Claude.

The widespread availability of generative artificial intelligence tools presents an important challenge to the integrity of documentary evidence in common law adversarial litigation. The courts are no strangers to fraudulently produced documents. But the arrival of large language models alongside text-to-image, text-to-audio, and text-to-video technologies has altered the threat. The barriers to creating a convincing forged document, once measured in terms of significant cost, skill, and time, have been minimised, if not eliminated entirely. Anyone with a smartphone can now produce a plausible fake in a matter of minutes at zero direct cost. Further, detection of the fake may require access to specialist tools. Are the procedural frameworks which govern disclosure across common law jurisdictions are adequate to meet that challenge?

Take two illustrative scenarios:

1. In relatively modest litigation, an AI-generated dashcam video (which is indistinguishable from genuine footage) could be deployed to misrepresent the facts of a minor road traffic accident. Innocent parties may find themselves pressured into settling, especially if expert evidence as to authenticity might be disproportionate to the sums at stake.
2. At the other end of the scale, a determined bad actor in complex commercial litigation might fabricate historical corporate records, constructed with period-appropriate language, authentic-sounding references, and plausible metadata, rendering verification extraordinarily difficult. The judgment in Crypto Open Patent Alliance v Wrightserves [2024] EWHC 1198 (Ch) provides a sobering reminder that the deliberate introduction of fabricated documents into legal proceedings is not a theoretical concern.

In England and Wales, CPR rule 32.19 establishes a presumption of authenticity for disclosed documents, which a receiving party may displace by serving a timely notice. Materially similar provisions exist under the Australian Federal Court Rules 2011 and the Singapore International Commercial Court Rules 2021. Once triggered, these mechanisms shift the burden of proof of authenticity onto the disclosing party. In principle, they provide a meaningful check on fraudulent disclosure.

However, these safeguards were designed for a world in which forging documents was costly. The presumption of authenticity was built on certain assumptions, including that (1) parties generally act honestly, (2) creating a convincing forgery is prohibitively difficult for most, and (3) the high risk of exposure acts as a deterrent. The second and third assumptions are now shaky, at best.

But abolishing the presumption of authenticity and requiring a party to prove the authenticity of each document it discloses would be wholly disproportionate in nearly all cases. Instead, consideration should be given to requiring parties and/or their legal advisors to certify that none of the documents disclosed had been created or materially altered by artificial intelligence. False or inaccurate certification would expose the relevant party to appropriate sanctions, creating a credible deterrent, without fundamentally restructuring the existing disclosure framework.