Practice Note: Use of Artificial Intelligence in Arbitration

NB: references to The ADR Centre include all subsidiaries, including the New Zealand Dispute Resolution Centre (NZDRC), New Zealand International Arbitration Centre (NZIAC), Building Disputes Tribunal (BDT), the Family Dispute Resolution Centre (FDRC), and the Independent Complaint and Review Authority (ICRA).

This Practice Note provides guidance on the use of artificial intelligence (AI) tools in The ADR Centre arbitrations.

It applies where a party, a representative, a witness, an expert, or the tribunal proposes to use AI tools in connection with documents, evidence, submissions, or hearing support.

This Practice Note does not override any agreement between the parties, any applicable law, or any procedural directions made by the tribunal, but is intended to support consistent, proper and principled conduct when AI tools are used.

1. AI Tools

AI tools include software that can generate, summarise, translate, classify, or analyse content, including generative AI systems.

AI tools can be useful. They can also be unreliable and may create risks for confidentiality, privilege, procedural fairness, and evidential integrity.

2. General Principle

Any use of AI tools in an arbitration must be consistent with:

• party equality and the right to be heard;
• confidentiality and privilege;
• accuracy and candour to the tribunal; and
• the integrity of the evidence and the [text cut off in source]

The party using an AI tool remains responsible for what is filed, said, or relied on.

3. Early case management

At an early stage, the tribunal may address whether an AI protocol is needed. The tribunal may invite the parties to identify:

• intended AI uses (for example, drafting assistance, document review, translation, expert analysis);
• whether confidential case material will be input to any AI tool;
• proposed safeguards; and
• any disclosure that is appropriate for [text cut off in source]

Any directions should be proportionate to the scale and sensitivity of the arbitration.

4. Confidentiality, privilege, and data security

Parties should not input confidential or privilege arbitration material into an AI tool unless they are satisfied that doing so will not compromise confidentiality or privilege.

Before using an AI tool with case material, parties should consider:

• whether the tool stores prompts and outputs;
• whether the tool uses inputs for training or improves its models;
• where data is processed and stored (including offshore locations);
• who can access the data (including subcontractors); and
• how data can be deleted [text cut off in source]

Where AI use involves confidential case material, the tribunal may direct safeguards, including:

• use of a controlled environment (for example, an enterprise/private instance);
• redaction or anonymisation;
• restrictions on categories of material (for example, personal data or trade secrets); and
• written confirmation of retention and deletion.

5. Accuracy and verification

AI tools can produce errors, including fabricated citations, incorrect quotations, and misleading summaries.

Any party relying on AI-assisted work must verify:

• all legal authorities and citations against the originals;
• quotations and pinpoint references;
• factual assertions against the evidential record; and
• translations or technical summaries where they are [text cut off in source]

If an error is discovered, the party should correct it promptly.

6. Disclosure and transparency

There is no automatic requirement to disclose routine or administrative AI assistance.

However, the tribunal may require disclosure where necessary for fairness, including where AI is used to:

• generate or materially revise evidence (including witness statements or expert reports);
• create, or enhance audio, video, images, or other digital evidence; or
• produce analyses that are relied on in [text cut off in source]

Any disclosure should be proportionate and may include:

• the tool used and its general purpose;
• whether confidential material was input (and at what level of sensitivity); and
• the verification steps [text cut off in source]

7. Witnesses and experts

Witnesses must give their own evidence.

The tribunal may direct measures to ensure that AI is not used to coach or prompt a witness during evidence, including:

• prohibiting real-time AI assistance during testimony; and
• confirming that no unauthorised communications or tools are used while the witness is giving evidence.

Experts remain responsible for their opinions and methodology.

Where an expert uses AI tools in forming an opinion, the tribunal may direct that the expert record:

• how the tool was used;
• what inputs and assumptions were applied; and
• what steps were taken to validate [text cut off in source]

8. Authenticity and manipulated media

AI tools can be used to create or manipulate digital material.

Where authenticity is in issue, the tribunal may direct steps such as:

• preserving metadata;
• providing chain of custody information; and
• arranging forensic examination where [text cut off in source]

Parties should promptly raise any reasonable concerns about manipulated or synthetic media.

9. Tribunal use of AI

The tribunal must personally discharge its adjudicative function. The tribunal must not delegate, outsource, or abdicate any aspect of decision-making to an AI tool. Any award, ruling, procedural order, or reasons must reflect the tribunal’s own independent analysis and conclusions.

Unless the parties expressly agree and the tribunal is satisfied it is appropriate (and lawful) in the circumstances, the tribunal must not use AI tools to:

• determine, recommend, or optimise the outcome of any issue in dispute;
• assess credibility, reliability, or weight of evidence;
• evaluate witness testimony or demeanour;
• make findings of fact or draw inferences from evidence;
• determine the content of procedural directions where those directions materially affect a party’s ability to present its case; or
• generate reasons that are adopted without full and independent [text cut off in source]

The tribunal may use AI tools for limited, non-adjudicative assistance, such as:

• formatting, proofreading, and style consistency;
• converting documents (for example, OCR, indexing, or compilation), where accuracy is checked;
• non-substantive summarisation of procedural history (for example, timelines of steps taken), where verified against the record; and
• administrative tasks that do not involve evaluating the [text cut off in source]

Where the tribunal uses any AI tool, it must ensure that confidentiality is preserved. The tribunal should not input confidential case materials into an AI tool unless satisfied that:

• the tool operates in a controlled environment (for example, an enterprise/private instance);
• prompts and outputs are not retained or used for training beyond the tribunal’s control; and
• appropriate access controls, retention, and deletion settings are in [text cut off in source]

If the tribunal uses AI for any purpose beyond purely administrative support, it should consider whether, and to what extent, disclosure to the parties is necessary to preserve confidence in the process and procedural fairness. Any disclosure should be proportionate and may include:

• the nature of the tool and general purpose of use;
• whether any case materials were input; and
• the verification steps [text cut off in source]

The tribunal must not rely on AI outputs as a substitute for reviewing the evidential record and submissions. Any AI-assisted summary, extract, or collation must be verified against the underlying documents and transcript. Any identified inaccuracies should be corrected promptly and not carried into reasons or orders.

The tribunal should avoid any AI use that could subtly shape or bias its reasoning (including “Argument ranking”, suggested findings, or “likely outcome” outputs). If any AI output might influence substantive analysis, the tribunal should not use it.

Where appropriate, the tribunal may adopt a short protocol (or procedural direction) recording:

• permitted tribunal uses (if any);
• an express statement that AI will not be used for decision-making;
• confidentiality and data-handling measures; and
• any disclosure [text cut off in source]

10. Further guidance

If a participant is uncertain about appropriate AI use in an arbitration, they should raise it with the tribunal at the earliest reasonable opportunity.

Issued by The ADR Centre. Effective from April 14, 2026.