CIArb Guidelines on AI – next step in the evolution of soft law on AI – Natalia Chumak and Neil Newing

Natalia and Neil's article was published in Law360, 7 May 2025, here.

Introduction to soft law on AI in arbitration

There is a growing body of soft law attempting to regulate, or rather, provide guidance on the use of AI in disputes and, specifically, international arbitration. CIArb's Guidelines on the Use of AI in Arbitration ("CIArb AI Guidelines"), which launched on 13 March 2025, is the latest and a welcome addition to the list. We consider how they compare to the existing soft law and whether they bring anything new to the table.

Soft law providing guidance on the use of AI in disputes first emerged at the end of 2023 when the American Arbitration Association’s International Centre for Dispute Resolution (AAA-ICDR) published its Principles Supporting the Use of AI in Alternative Dispute Resolution (November 2023). This is a short document which set out basic and very high-level principles of integrating AI into ADR. Other institutions followed suit with the Silicon Valley Arbitration & Mediation Center (SVAMC) publishing their AI Guidelines in April 2024 and SCC Arbitration Institute publishing a Guide to the use of AI in cases administered under the SCC rules later that year in October.

Whilst SVAMC’s AI Guidelines are the most detailed of the three, they have all adopted the same approach, providing a high-level principle-based framework for the parties and arbitrators to use as a checklist when considering the use of AI in arbitration. Such level of guidance is a useful starting point, but as the use of AI becomes more and more a part of everyday professional life, parties and arbitrators are likely to require a more detailed instrument setting out more detailed and practical guidance on how to use AI tools.

The CIArb AI Guidelines represent that next step in the evolution of soft law regulation. It is a more elaborate document which seeks to provide practical guidance on how to manage the risks of AI within an arbitration process, while reinforcing the importance of maintaining party autonomy, a key principle of arbitration.

CIArb AI Guidelines – structure

CIArb AI Guidelines are split into four parts, with two appendices:

• Part I: Benefits and risks of the use of AI in arbitration
• Part II: Recommendations on the use of AI in arbitration
• Part III: Arbitrators’ powers to give directions / make rulings on the use of AI by parties in arbitration
• Part IV: Use of AI in arbitration by arbitrators
• Appendix A: Template Agreement on the Use of AI in Arbitration
• Appendix B: Template Procedural Order on the Use of AI in Arbitration

The most innovative sections are Parts III, IV and the Appendices. 

Part III of the CIArb AI Guidelines – Parties' Use of AI in Arbitration

The most notable feature of Part III is that arbitrators are encouraged to take a proactive and wide-ranging role in managing the use of AI in arbitrations. This includes addressing the use of AI from the outset in the proceedings and recording any decision made in a procedural order; appointing AI experts to understand the relevant AI tool and the consequence of its use better; requiring disclosure of AI use; addressing the use of AI in its award; taking into account in its costs award any failure by the parties to comply with the tribunal's directions regarding the use of AI.

Part III of the CIArb AI Guidelines - specific provisions on arbitrators' powers to rule on admissibility of AI-generated material (paragraph 6)

In relation to admissibility paragraph 6.2 provides "If the arbitrators consider that the use (or non-use) of AI by one or more party jeopardises the integrity of the arbitral proceedings, arbitrators may make a ruling on the use of AI of their own motion after consulting the parties."

Paragraphs 6.4 – 6.6 include suggestions of the issues arbitrators may consider or information an arbitrator may request the parties to provide in order to allow them to make a ruling on the use of AI.

Whilst the suggestions are undoubtedly sensible, they assume that arbitrators have/will have a high level of AI literacy, even if they are able to use an AI expert in order to assist them with their assessment. The provisions also imply that the parties will have sufficient information about the structure of the AI Tool in question and the data that has been used to train it. This may be challenging since some AI tech companies are unlikely to be sufficiently transparent about their know-how and the data used for training.

Paragraphs 6.7 – 6.8 provide that the arbitrators must consider applicable laws related to the use of AI which includes a) the law of the seat of arbitration; b) the laws and rules governing the proceedings; c) the national laws of the parties; and d) any applicable ethical rules. Such laws may not be directly aimed at regulating AI but may be indirectly relevant to the decision on the use of AI in arbitration.

It is likely that the arbitrators will not always have sufficient expertise in AI-related laws in the relevant jurisdictions in order to decide on the effect these laws have on the use of AI. This means that the arbitrators will require the parties to make submissions and/or provide expert evidence on the relevant applicable laws.

Part III of the CIArb AI Guidelines - specific provisions on arbitrators' powers to rule on disclosure of the use of an AI Tool (paragraph 7)

Paragraph 7 sets out the parameters in which disclosure may be required and, again, encourages the arbitrators to be proactive in managing this issue. Paragraph 7.3 gives arbitrators powers "Unless otherwise agreed by the parties, and after consulting them, … to impose certain AI-related disclosure obligations on the parties…"

In comparison, the starting position in SVAMC AI Guideline 3 is that disclosure is not necessary. Whilst the commentary to SVAMC AI Guideline 3 acknowledges the possibility of disclosure and adds some details as to when disclosure may be appropriate, there is no suggestion that arbitrators should take a proactive role on this issue.

Part IV of the CIArb AI Guidelines – Use of AI in arbitrations by arbitrators

Like SVAMC AI Guidelines, Part IV of the CIArb AI Guidelines encourages arbitrators to consult with the parties on the use of any AI Tool and provide them with an opportunity to comment. However, CIArb AI Guidelines go a step further by saying that the arbitrators "should" (emphasis added) refrain from using the specified AI Tool. This provision stops short of imposing an obligation on arbitrators to refrain from using AI-generated information if the parties disagree with such use.

Appendices of the CIArb AI Guidelines

Appendix A contains a template agreement on the use of AI in arbitration which can be used as a standalone agreement or as part of an arbitration agreement. It is worth noting that if such an agreement is incorporated into the substantive contract between the parties, which may be concluded years before a dispute arises, the approach adopted in the agreement on the use of AI may be out of date by the time the dispute has arisen.

Appendix B contains two versions (a short one and a long one) of a procedural order on the use of AI in arbitration.

Even if the templates contained in the Appendices are not adopted in their original form, they serve as a useful starting point for the parties and for the arbitrators who are considering how to regulate the use of AI. These are more than other Guidelines have provided to date and are a step in the right direction.

Conclusion

The CIArb AI Guidelines have highlighted several issues which the arbitration world will have to address in the near future as part of getting used to, and making use of, the new tech tools. First, there is a great and urgent need for a high level of AI literacy for all stakeholders in the arbitration process. The credibility and the integrity of the arbitral process depends on the ability of the parties and the arbitrators to stress-test AI Tools and AI-generated information which is only possible with in-depth knowledge of how AI Tools work.

Second, there may be an increase in costs and delays in some arbitral proceedings as a result of having to seek expert evidence on AI-related issues and clarifications on applicable mandatory AI-related laws. These issues are likely to settle down in time once the industry forms established practices of handling AI-related matters.

Finally, and more generally, AI tools are developing fast, and the experience of their use by arbitration practitioners is also evolving rapidly to keep pace. It has been suggested by some commentators that since this rapid development is likely to make any guidance issued obsolete quickly, the attempts to provide such guidance are therefore futile.

However, there is an alternate view to this. Rapid evolution of AI's presence in the legal industry, and in international arbitration in particular, makes it imperative for the main stakeholders (ie industry bodies, arbitral institutions, leading practitioners) to respond - and continue responding - to these developments. They simply need to do it quickly.

Prompt, transparent and collaborative thought leadership is required to address the challenges and opportunities created by AI. In our view it is the only way forward to preserve the credibility of the arbitral process and, as part of that, to reduce the risk of challenges to arbitral awards (for example, if AI were to be used by parties and/or arbitrators without proper transparency and agreement over its use, and it is subsequently discovered that authorities relied on or findings of fact were made based on AI hallucinations).  The CIArb AI Guidelines is an example of this approach to try to ensure that the parties and the arbitrators are not only aware of the relevant issues surrounding the use of AI but are actively addressing them in the arbitration process and, ultimately, if they are unable to agree on how AI is to be used, to ensure that the arbitrators are able to rule on such issues so all parties are clear in this regard. However, the soft law approach is not devoid of challenges. There is an issue of likely inconsistency in interpretation, and therefore application, of soft law by the parties and the arbitrators, and where there is disagreement such that a ruling is required, that will place a heavy burden on arbitrators to get that right as a ruling which improperly permits (or restricts, although it is more likely the former) the use of AI could still lead to an award being challenged. Given that arbitral awards are typically confidential, it may take a long time for soft law to develop into reliable and authoritative guidance which will be applied uniformly. As such, it may well be that while we are still in the early stages of considering how AI can be adopted into arbitration matters, and particularly where there is disagreement between the parties in this regard, that the more common approach for now will simply be to take a cautious approach to its use.