AI in the Arbitration Room — Threat or Tool for Emerging Institutions?

There is a moment in every significant technological shift, when the legal profession stops watching from the sidelines and is forced to actually engage. We are in that moment right now with artificial intelligence and international arbitration. The question is no longer whether AI will enter the arbitration room. It already has. The more pressing question, particularly for smaller and emerging arbitral institutions trying to find their place in a competitive global landscape, is whether this changes the game in their favour or makes the climb considerably steeper.

I want to be honest about something from the start. When I first began reading about AI in arbitration, I thought the conversation would be dominated by the obvious efficiency arguments like faster document review, cheaper proceedings and AI-assisted research. And while those arguments are real and worth taking seriously, I think they miss the more complicated story that is actually unfolding. The story is not just about tools. It is about power, trust, and who gets to shape the future of international dispute resolution.

What Is Actually Happening Right Now?

A lot has happened in the last year and a half. In March 2025, the Chartered Institute of Arbitrators came out with its Guidelines on AI in Arbitration. It basically tells arbitrators to actively manage how AI is used — and to disclose it when AI plays a big role in proceedings. Around the same time, the Stockholm Chamber, the Vienna Centre, and the Silicon Valley Center all put out their own guidance. The message is the same everywhere: AI is here, use it carefully, don’t let it replace human judgment. 

Then in November 2025, something more dramatic happened. The American Arbitration Association and its International Centre for Dispute Resolution announced the deployment of an AI arbitrator for documents-only construction cases, the first time a major institution had actually allowed an AI system to generate draft awards in real disputes. The AI was trained on over 1,500 construction arbitration awards, developed with input from experienced practitioners, and operates under a human-in-the-loop framework where human arbitrators review and, if needed, revise the AI’s output before finalisation. Both parties must consent. The AAA-ICDR described it as, in their own words, “probably the most exciting thing we’ve announced in the last 100 years.”

That is either a thrilling sign of things to come or a slightly alarming one, depending on where you sit.

The Case for AI as a Tool

Let me take the optimistic view seriously first, because it deserves to be taken seriously.

The biggest structural problem in international arbitration is cost and delay. Complex commercial disputes routinely take three to five years and cost millions in legal fees. Document review alone is the process of going through thousands of contracts, emails, and financial records to identify relevant evidence , is extraordinarily time-consuming and expensive when done by human lawyers billing by the hour. AI changes this calculation dramatically. Tools like Harvey and Clearbrief are already being used by major law firms to go through submissions, find relevant case law, and catch inconsistencies in documents, work that used to take weeks now takes hours.

For emerging arbitral institutions, those based in Southeast Asia, the Middle East, Africa, and South Asia that are competing for cases against the established giants in London, Paris, and Singapore — this is potentially genuinely transformative. If AI can bring the cost of running proceedings down significantly, suddenly choosing a newer institution over London or Paris makes a lot more financial sense. If AI can assist in producing consistent, high-quality procedural orders and case management decisions, it partially compensates for the relative inexperience of newer institutions compared to the ICC or LCIA who have been doing this for decades.

There is also the access to justice argument, which I think gets underplayed in a lot of the academic commentary. A significant proportion of commercial disputes, particularly in developing markets , never make it to arbitration at all because the cost is prohibitive. If AI-assisted arbitration can bring the price of resolving a USD 50,000 dispute down to something a mid-sized business can actually afford, that is a genuine social good. The AAA-ICDR’s AI arbitrator, limited initially to lower-value construction cases, is essentially an experiment in exactly this direction.

The Complications

Now for the honest part.

The concerns around AI in arbitration are not minor procedural quibbles. They go to the core of what arbitration is supposed to be. Arbitration works because people believe in it, because they trust that the process was fair and that a real human being with genuine judgment heard their side. Parties accept the outcome because they believe they had a fair opportunity to present their case to an impartial, qualified human being who applied genuine judgment to the evidence. The moment you introduce an AI system into that chain, even in a supporting role — you are adding a layer that neither party fully understands and that neither can meaningfully interrogate.

The one issue I keep coming back to is confidentiality. International arbitration proceedings are private by design. Documents submitted include commercially sensitive contracts, financial records, trade secrets, and strategic information that parties would never share publicly. When AI tools process this material, where does the data go? Who has access to it? The ICDR and SIAC Rules 2025 both address confidentiality, but the reality is that most AI tools, even enterprise versions , involve some level of data transmission to external servers. For parties dealing with genuinely sensitive disputes, this is a real and legitimate worry.

Then there is the hallucination problem, and it is a real one. AI systems regularly produce legal analysis that sounds completely authoritative but is just wrong. Citing cases that do not exist. Applying the right legal principle to completely the wrong situation. In a draft arbitral award, that is not a minor editing issue. That is a challenge to enforcement. In a legal brief, a hallucination caught by a careful reviewer is an embarrassment. In a draft arbitral award, it could be grounds for challenge and could potentially undermine enforcement. The EU AI Act classifies AI decision-making in legal proceedings as high-risk for exactly this reason.

And there is a deeper concern that I think deserves more attention than it usually gets: the question of cultural and contextual judgment. International arbitration increasingly involves parties from very different legal and commercial cultures, disputes where the background relationships are as important as the documents, and situations where the equities of a case, what is actually fair given the full context , require the kind of nuanced human judgment that no AI system currently possesses. An AI arbitrator trained predominantly on Western commercial arbitration awards is likely to embed Western legal assumptions and reasoning patterns in ways that may not serve parties from different systems fairly.

What This Means for Emerging Institutions Specifically?

Here is where I think the debate gets particularly interesting and, frankly, not talked about enough.

The established institutions like ICC, SIAC, LCIA, have the brand recognition, the case volume, the experienced staff, and the financial resources to invest in AI tools and develop proper governance frameworks around them. SIAC’s 2025 Rules already incorporate mechanisms that work alongside new technology rather than against it. These institutions can afford to experiment cautiously.

Emerging institutions face a more difficult calculation. The upside is obvious: lower costs, better consistency, more cases handled with leaner teams. But the downside is also real. Any significant AI-related failure, like a confidentiality breach, a poorly reasoned AI-assisted award that gets challenged, a hallucination that embarrasses the institution, could be devastating for a body that is still trying to establish its reputation.

There is also the regulatory landscape to navigate. The EU AI Act, which came into force in August 2025, classifies AI involvement in legal proceedings as high risk and imposes significant transparency and oversight requirements. Any institution with European parties or European-seated proceedings will need to think carefully about compliance. This is a burden that well-resourced institutions can manage more easily than smaller ones.

If I had to give one practical piece of advice to emerging institutions, it would be this: do not jump into AI-assisted decision-making right away. Start with the low-risk stuff. Use AI for document management, case tracking, helping case managers with research, and translation support in multilingual cases. These things do not carry the same risk as having AI help write awards, and they can still save you a lot of time and money.  Second, watch what happens with the AAA-ICDR AI arbitrator pilot over the next two to three years before making any decisions about AI-assisted award generation. That experiment will generate real data on enforcement challenges, party satisfaction, and the practical limits of AI judgment in dispute resolution.

The Honest Conclusion

I started writing this piece fairly convinced that the answer to the title question was straightforward — AI is a tool, clearly, of course it is. By the time I finished researching it I am genuinely less certain. The efficiency gains are real. The access to justice benefits are real. But the risks to due process, confidentiality, and the fundamental trust on which arbitration depends are also real, and they are not going to be solved by guidelines alone.

What I am sure of is this: the institutions that will benefit most from AI are the ones that approach it with genuine intellectual seriousness rather than either uncritical enthusiasm or reflexive resistance. For emerging institutions in particular, the goal should not be to compete with the ICC by replicating what the ICC does. The goal should be to use technology — AI included — to build something genuinely better: faster, cheaper, more accessible, and equally trustworthy. That is a harder task than it sounds. But it is the right one.