AI Disputes in Arbitration: How the JAMS AI Rules Streamline eDiscovery in ADR

Even before AI, the identification, collection, and review of electronically stored information had outgrown the patience of most discovery frameworks. AI systems make the problem worse in specific, predictable ways. They generate large volumes of unstructured output — chat logs, sensor readings, model inferences — that does not sit neatly in the rows and columns discovery tools were built around. The models themselves change as they learn, so the system that produced a disputed output may no longer exist by the time anyone goes looking for it. And many models are effectively black boxes, which means the fight is rarely about a document; it is about why a system did what it did. That is a different kind of discovery, and litigation's default machinery is poorly suited to it.

Arbitration handles this better than most people expect, and the JAMS AI Rules were written for exactly these disputes. The advantages are not about cutting corners on transparency — ADR parties increasingly demand the same data transparency as court litigation — but about putting the right decision-maker and the right procedures in front of a technically hard problem from the start.

Why AI evidence breaks ordinary discovery

It helps to be concrete about what makes AI-related ESI difficult, because the difficulties are what the right procedure has to address.

• Volume and variety. AI systems produce unstructured data that resists ordinary categorization, so the population to be collected and reviewed is both larger and messier than a typical email corpus.
• Reproducibility. A model that keeps learning may not reproduce the output at issue, which puts a premium on snapshots, audit trails, and configuration history rather than the live system.
• Interpretability. When liability turns on how a model decided something — as in a discrimination or product claim — discovery may have to reach the model architecture, training datasets, and configuration, not just the outputs.
• Preservation and spoliation. Continuous learning and retraining can overwrite the relevant state of a system, so preservation has to be scoped to model parameters and training data, not only files.
• Privacy and proprietary data. AI systems ingest data from many sources, some subject to GDPR, the CCPA, or trade-secret protection, which layers compliance obligations onto every collection decision.

Each of these is a question a generalist judge with a crowded docket is not well positioned to resolve quickly. That is the gap arbitration can fill.

What the JAMS AI Rules actually change

The JAMS AI Dispute Resolution Rules were designed for the realities of AI technology rather than retrofitted from a generic discovery regime. Three features matter most for managing eDiscovery.

A technically competent decision-maker

The single biggest advantage of arbitration here is the ability to select an arbitrator — or a discovery referee — who actually understands AI systems and eDiscovery. Most judges do not, through no fault of their own; the subject matter is specialized and the law is still developing. An arbitrator who can follow an argument about search terms, sampling, model explainability, or data provenance resolves scope disputes faster and more sensibly, and is better positioned to rule on the merits in an area where precedent is thin.

Procedural flexibility and proportionality

The rules let the parties and the arbitrator tailor discovery protocols to the specific technology in dispute, and they empower the arbitrator to narrow overbroad or marginally relevant requests. That proportionality discipline is exactly what opaque, evolving systems demand: the burden of producing model internals or full training sets has to be weighed against what the requesting party genuinely needs to prove its case. Calibrating that balance early keeps the discovery from swallowing the dispute.

Neutral experts and confidentiality

The framework contemplates appointing neutral experts who understand both the legal and the technical sides of AI, which speeds the review of complex digital evidence and reduces the risk that a tribunal misreads what a model did. Arbitration also offers stronger, more practical confidentiality protections than open court — a real concern when the evidence is proprietary source code, model weights, or training data a party cannot afford to expose.

Practical guidance for practitioners

Whether you are new to ADR or simply new to AI evidence, the work of managing it well is concrete. Competence with this technology is not optional — ABA Model Rule 1.1, Comment 8 frames keeping abreast of relevant technology as part of the duty of competence — but it is learnable. A few priorities carry most of the weight.

1. 01Scope preservation to the system, not just the files. Identify model versions, training data, and configuration early, and capture a defensible snapshot before continuous learning overwrites the state at issue.
2. 02Negotiate the AI discovery protocol up front. Agree on which sources are in play — model internals, logs, training sets — and on proportionality limits before a dispute hardens, rather than litigating scope mid-arbitration.
3. 03Press for a technically capable neutral. Where the evidence turns on how a model behaved, ask early about an arbitrator or special master with AI and eDiscovery fluency, and consider a neutral expert to interpret competing analyses.
4. 04Keep human judgment over AI-assisted review. Predictive coding and generative tools speed review, but they can embed bias from a skewed training set and invite under-inclusivity arguments; validate the output and do not delegate consequential calls to the model.
5. 05Be transparent about AI use. ADR runs on trust, so disclose where AI tools are deployed in your own review or analysis, and confirm that processing complies with applicable privacy regimes and any protective order.
6. 06Collaborate. Work with forensic experts, IT, and opposing counsel to set fair discovery parameters — the technical nature of this evidence rewards cooperation and punishes posturing.

Where AI helps the process itself

The same technology driving these disputes can also streamline them. AI-powered review tools sift large volumes of ESI faster than manual review, and predictive coding lets a model learn from human coding decisions and apply that to the broader set — a meaningful efficiency in ADR, where parties usually want a faster, cheaper resolution. Used carefully, AI can even help a tribunal by synthesizing dense technical material into something digestible. None of that displaces the validation discipline; if anything, a model whose reasoning a human cannot fully audit raises the bar for proving the result is reliable.

The bottom line

Arbitration is not a way to do less discovery in an AI dispute; it is a way to do the right discovery in front of someone equipped to manage it. The JAMS AI Rules supply the structure — technical competence at the decision-maker level, procedural flexibility, proportionality, neutral expertise, and confidentiality — that opaque and fast-moving AI systems require. For practitioners, the payoff comes from engaging the technical questions early rather than after a production goes sideways. If your matter involves an AI system, a tech-heavy arbitration, or a discovery dispute that needs a neutral who can read the technology as well as the record, you can start a scoping conversation through our home page or email the team directly to discuss a conflict check and approach.