By Javier Surasky
Artificial intelligence is already part of dispute resolution, and its use is expanding across arbitration, mediation, and other digital mechanisms. The central question is no longer whether AI should be used, but under what conditions it can be integrated without undermining confidentiality, due process, equality between the parties, and human responsibility for decisions.
Current Uses and UNCITRAL’s Work
The current
range of AI uses is broad: the selection of arbitrators or mediators, legal
research, gathering documentary evidence, document review, outcome prediction,
quantification of damages, translation and transcription, and case management.
In practical terms, this means that it can help process large volumes of
information, detect patterns, organize case files, prepare timelines, or reduce
administrative burdens.
This
potential led the United Nations Commission on International Trade Law
(UNCITRAL) to launch the DRDE project in 2021, aimed at monitoring changes in
dispute resolution in the digital economy. Within that framework, in September
2025, its Secretariat consulted States on possible lines of work concerning AI,
digital platforms, and remote proceedings.
The
responses show a clear preference for concentrating the use of AI on auxiliary
functions such as case management, gathering documentary evidence, and legal
research, but also widespread reluctance to apply it to outcome prediction, the
selection of arbitrators or mediators, or granting it decision-making capacity.
The main
boundary, then, appears to lie in the division between assisting and deciding,
and it is reflected in documents prepared by expert institutions: the Silicon
Valley Arbitration & Mediation Center, the Chartered Institute of
Arbitrators, and the Vienna International Arbitral Centre have published
guidelines or notes on AI in arbitration and, although their approaches are not
identical, they converge on the idea that AI can support the conduct of
proceedings, the organization of information, or certain analytical tasks,
provided that it does not replace the arbitrator’s independent judgment.
Transparency, Confidentiality, and Human Decision-Making
A first
element behind this lies in the transparency challenge affecting AI: the
Chartered Institute of Arbitrators warns about the “black box” problem, which
makes it difficult, if not impossible, to understand how AI reaches certain
results, leaving an area of opacity that affects the possibility of
understanding and verifying the accuracy of its statements.
That risk
becomes worse when errors, biases, or “hallucinations” appear: a misquoted
legal citation, the omission of a document, or the incorrect classification of
evidence may affect the rights of the parties and, when the dispute is
international or involves parties from different cultures, the risk increases
exponentially.
Another
critical point raised is confidentiality, since disputes may involve sensitive
information, personal data, and protected communications that, when entered
into an AI tool, trigger issues related to their storage, reuse, use for model
training, unauthorized access, and data transfers. In fact, the Silicon Valley
Arbitration & Mediation Center is categorical on this point when it
recommends not submitting confidential information to AI tools without
appropriate assessment and authorization.
Somewhat
less precise, but in the same vein, the IBA guide for mediation stresses that
users must take reasonable measures to ensure that confidential information is
not compromised by the use of AI, especially when proprietary or open-source
models are used.
Together,
these concerns explain why decision-making authority must remain in the hands
of the persons appointed for that purpose.
The
Chartered Institute of Arbitrators adopts the perspective of procedural risks,
emphasizing that the use of AI may affect due process, equality between the
parties, confidentiality, data security, and the enforceability of the award,
and therefore maintains that arbitrators must retain responsibility for the
outcome and use their own reasoning to justify their decisions.
The Silicon
Valley Arbitration & Mediation Center makes the strongest statement, saying
that “an arbitrator shall not delegate any part of their personal mandate to
any AI tool.”
The Vienna
International Arbitral Centre states that arbitrators must retain full control
over the decision-making process and must not delegate to AI any decision that
could have an impact on it.
The most
salient risk is AI’s capacity to influence decisions without formally replacing
the arbitrator. For that reason, human control over proceedings should not be
understood as a review of the result produced by AI, no matter how expert it
may be, but as the effective responsibility to apply one’s own reasoning
throughout the entire process and, especially, when considering the evidence
and adopting the final decision.
Mediation and Arbitration: Distinct Risks.
In
mediation cases, all these risks take on a particular form, since mediation is
a dispute resolution mechanism based on party autonomy and therefore
presupposes sufficient levels of trust and a willingness to reach settlement
agreements and implement them. Yet the impact of the problems identified may
end up weakening both.
In
addition, mediation depends on factors that are difficult to automate, such as
reading power dynamics, managing caucuses (confidential sessions between the
mediator and only one of the parties), cultural sensitivity, and the
progressive strengthening of trust in the achievement of a fair and mutually
acceptable solution.
For that
reason, AI becomes problematic if it is given the opportunity to decisively
shape the interaction between the parties.
In
arbitration, the debate extends to the validity and enforcement of the award.
If a party
argues that AI improperly influenced the decision, for example by using
information outside the case file or by acting in a way that broke procedural
equality between the parties, it could raise objections linked to due process,
a concern that “several States” pointed out in their communications to
UNCITRAL, according to its report A/CN.9/1271, paragraph 14.
Closing remarks
The
emerging tension is between an extremely limited application of AI, which would
prevent legitimate and efficient uses, and the absence of minimum common
criteria, which would end up weakening trust in the processes.
To address
it, professional guidelines propose flexible principles, not always with
properly defined contours, relying on concepts such as technical competence,
due diligence, or responsibility for outcomes, among others, along with the
already mentioned prohibition on delegating decision-making functions.
That
orientation coincides with UNCITRAL’s approach. Its Working Group on dispute
resolution issues observed that AI is a relevant and rapidly evolving area and,
before moving toward uniform rules, considered it preferable to continue
monitoring in order to identify related problems.
Even so,
minimum points of consensus already exist: AI must be understandable in its
relevant uses, secure in its handling of information, and compatible with
equality between the parties, and decisions must always remain under the
control and judgment of qualified individuals.
AI is
already a tool present in dispute resolution, but there is still no single international framework organizing its uses. The challenge lies in designing
practices that make it possible to take advantage of its efficiency without
turning arbitration or mediation into opaque, depersonalized processes put
beyond the control of those responsible for conducting them.
The
technological promise is valuable, but it will bear fruit only if it
strengthens trust in dispute resolution and reinforces the values of
independence, impartiality, confidentiality, due process, and, above all, human
responsibility.