Disclosure of AI use in arbitration: who must disclose what, and when

The short version

• No law or institutional rule we checked (how we search) requires AI disclosure in every arbitration.
• For arbitrators, Guideline 7 of the Silicon Valley Arbitration & Mediation Center (SVAMC), where adopted, and, for generative AI, California law from 1 January 2027 call for disclosure before relying on AI-generated information outside the record.
• For parties, the Chartered Institute of Arbitrators (Ciarb) lets the tribunal order disclosure, including from experts and fact witnesses.
• Four first procedural orders we read, made by tribunals in International Centre for Settlement of Investment Disputes (ICSID) cases, deal with AI. None makes parties disclose up front.
• No court we found has set aside an award because AI use was not disclosed.^{1,2,23,24,25,26,28,29,40,41}

Must AI use be disclosed in arbitration?

It depends on which text applies: no law or institutional rule we checked requires it in every case, and most AI texts bind only where adopted. Our guide to AI in arbitration compares the same texts on who decides. Here, AI disclosure in arbitration means telling other participants that AI was used; document production and deepfake evidence are in AI document review and deepfake evidence.^1,8,14

In our grouping, the texts use four triggers (AI rules tracker):

• Reliance on AI-generated information from outside the record (SVAMC; California).
• An effect on the evidence, outcome or reasoning (Ciarb; the American Arbitration Association and International Centre for Dispute Resolution, AAA-ICDR).
• The task the AI does (the SCC Arbitration Institute; the National Academy of Arbitrators, NAA).
• The tribunal’s discretion (the Vienna International Arbitral Centre, VIAC; the WIPO Arbitration and Mediation Center).^{1,2,4,7,8,10,12,28}

The LCIA’s rules consultation lists “AI disclosure and guardrails” as a theme; we found no disclosure terms from ICC’s AI task force. Our explainers cover ICC, the LCIA, AAA-ICDR, JAMS and SIAC.^18,19

When must an arbitrator disclose their own use of AI?

An arbitrator should disclose before relying on AI-generated information that is not in the record. SVAMC Guideline 7, where adopted, bars such reliance “without making appropriate disclosures to the parties beforehand” and, as far as practical, lets the parties comment (p 12); a footnote keeps iura novit arbiter, the arbitrator’s power to apply law the parties did not cite (fn 3). California’s section 1282.1(b)(2) is worded close to Guideline 7 but covers generative AI only, from 1 January 2027. Our explainer sets out the SVAMC Guidelines rule by rule.^1,28,29

AAA-ICDR’s January 2026 guidance says arbitrators “should disclose” generative AI use that “materially impacts” the process or reasoning, such as outputs that “inform or influence substantive reasoning, award language, or analysis of evidence”. Grammar checks, formatting and AI built into the AAA-ICDR case platform are excluded (Consideration 4).^3,4

NAA Advisory Opinion No. 28 sets the widest arbitrator duty to disclose AI use that we found. Labour-management arbitrators “must disclose” generative AI use in the substantive management of a case, such as summarising transcripts or drafting an award (Part II).¹²

Ciarb encourages arbitrators to consult the parties before using AI and to refrain if they disagree (para 9.1), while its template agreement says the tribunal “shall disclose the use of any AI Tools in advance of the relevant use” (Appendix A para 10.2.1). The NAA requires both parties’ consent before AI helps with “any decision-making function” (Part IV), and AAA’s AI-led process needs both parties to opt in (R-2(c)). Our explainer covers the Ciarb Guideline part by part.^2,6,12

Non-delegation applies whatever is disclosed (SVAMC Guideline 6; section 1282.1(b)(1)). What arbitrators may draft with AI is in arbitrators drafting awards with AI.^1,28

Do parties and their lawyers have to disclose AI use?

Usually not, unless the tribunal orders it. No rule or law we checked makes parties disclose AI use in arbitration by default. SVAMC Guideline 3 leaves it case by case (p 10), and its commentary notes that professional rules may require it (p 17).¹

Ciarb’s AI disclosure provisions are the clearest. Subject to the parties’ private use of AI, the tribunal “may require disclosure” (para 4.4) where use may affect the evidence or outcome or involves delegating a duty (para 7.1), and after consulting the parties it sets the scope, recipients and deadline (para 7.3). VIAC (para 5.1), CIETAC (paras 3.1, 5.2) and WIPO (section 5) leave it to agreement or the tribunal, and AAA’s AAAi Standards ask advocates to explain their AI use (Standard 4).^2,5,8,9,10

California’s rule on counsel AI disclosure in submissions covers courts only. From 1 January 2027, Business and Professions Code section 6068.1(a)(3)(C) requires lawyers to disclose generative AI use “to the court for all documents submitted to the court”. Court rules are in AI in law, and lawyers’ duties in AI legal tech.^28,29

Do experts and witnesses have to disclose AI use?

Only if the tribunal orders it or the parties agree. No text we checked sets a default rule on expert witness AI disclosure in arbitration. Under Ciarb, the tribunal’s power reaches “any party-appointed experts or factual witness” (para 7.3), and its longer template order treats an expert report as a party submission when “High Risk AI Use” (use that risks confidentiality, data security, procedural integrity or a “non-human influence on the award”) is disclosed, giving the other side a suggested seven days to comment (Appendix B, Option 2, para 3).²

The ADR Centre lets the tribunal ask an expert to record how a tool was used and the inputs and assumptions applied, and bar real-time AI help during testimony (section 7). How AI transcription and interpretation affect the hearing record is in AI transcription and interpretation at hearings.¹⁵

The International Bar Association (IBA) Rules on the Taking of Evidence, which apply where the parties agree or the tribunal decides (art. 1.1), do not mention AI. In our reading they support expert report AI disclosure: a report must describe “the methods, evidence and information used” (art. 5.2(e)).³⁹

What should an AI disclosure say, to whom and when?

At least the tool and what it was used for, given to whoever the tribunal directs (Ciarb para 7.3). A disclosure duty the tribunal imposes runs through the whole case unless it orders otherwise (para 7.4).²

• SVAMC Guideline 3, where appropriate, as details that “may help reproduce and evaluate the output”: the tool’s “name, version, and relevant settings”, a short description of its use, and “the complete prompt”, including the conversation thread, with the output (p 10).
• Ciarb’s longer template order, for High Risk AI Use: the tool, the document or activity, any passages generative AI drafted, and a certificate that no confidential information reached an unauthorised party (Appendix B, Option 2, para 2).
• Ciarb Appendix A para 8 and ARB/25/42 s 28.3: where the tribunal requires it, parties help reproduce outputs “by disclosing the prompts”.^1,2,25

For arbitrators’ own use, SVAMC says “beforehand”, AAA-ICDR “as soon as practicable” and the NAA “whichever is earliest” of planned or actual use; for other participants, Ciarb’s longer order says “promptly” (Appendix B, Option 2, para 2).^1,2,4,12

Whether AI prompts are privileged in arbitration is open. SVAMC weighs “any applicable privilege” and Ciarb any “legal impediment” (para 7.5), but no text we checked decides it. An AI disclosure statement template for mediation, from the IBA Mediation Committee (Part three), is covered in AI in dispute resolution. Keeping case material confidential when AI is used is a separate question, covered in confidentiality and AI in arbitration.^1,2,50

How do you deal with AI disclosure in Procedural Order No. 1 or terms of reference?

Raise it at the first case management conference or preliminary hearing, and record the result in Procedural Order No. 1 (PO1) or, where used, in terms of reference, the document setting out the claims and issues (VIAC paras 5.1, 5.3; Ciarb paras 4.6, 5.3 and the Appendix B note; KCAB art. 16.5; AAA-ICDR Consideration 2). The ICC Rules 2026 have no AI provision and do not require terms of reference; ICC’s 1 June 2026 Note lists them as an option where the parties and the tribunal decide it is appropriate (Part III para 6(h)).^{2,4,8,11,16,17}

• SVAMC’s model clause applies its Guidelines “as guiding principles” to all participants and adds no disclosure duty (p 21).
• Ciarb’s template order: Option 1 says all are “guided by” the Guideline; Option 2 requires prompt written disclosure of High Risk AI Use and lets the parties comment before the tribunal’s own high-risk use (Appendix B, paras 2 to 4).
• AAA-ICDR’s sample order: the arbitrator “will disclose any substantive use of AI tools that materially informs its reasoning or findings” (Consideration 4).^1,2,4

In the four ICSID orders below, starting with ICSID Case No. ARB/24/46, no AI clause makes parties disclose up front, and the ARB/25/24 order of 25 February 2026 has no AI section. In our reading, the party duties in ARB/25/42 s 28.2 and ARB/25/43 s 29.2 repeat Ciarb’s template agreement nearly verbatim (Appendix A paras 7.5 to 7.7), as ARB/25/42 s 28.3 does para 8.^{2,23,24,25,26,27}

For the points an AI clause should settle, with model clause and procedural-order wording, see AI arbitration clauses.

Does any law require AI disclosure in arbitration?

California’s law will, from 1 January 2027 (California Constitution, art. IV, s 8(c)(1)). SB 574, approved and filed on 30 September 2026 as Chapter 858, adds Code of Civil Procedure section 1282.1(b)(2): an arbitrator may not rely on generative AI information outside the record “without making appropriate disclosures to the parties beforehand and, as far as practical, allowing the parties to comment on its use”. We found no other statute aimed at arbitrators’ own AI use, though EU law will require some users of AI to tell the people affected.^28,29,45

H.R. 9715, a federal bill introduced on 15 July 2026, would in its introduced form allow AI in arbitrations under title 9 of the US Code only with each party’s written consent, and bar generative AI (proposed 9 U.S.C. section 2a(b), (d)).³⁰

Under the EU AI Act, from 2 December 2027 (Regulation (EU) 2026/1744, art. 1 point (40)(b)), the deployer of a high-risk Annex III system, which can include AI used in alternative dispute resolution (ADR), must tell the people it is used to decide about (art. 26(11)). Consumer ADR bodies using automated decision-making must tell parties in advance, under national rules applying from 20 September 2028 (Directive (EU) 2025/2647, art. 1 point (4)(b), art. 5). See AI in arbitration and AI in dispute resolution in Europe.^45,46,47

We found no official Peruvian rule requiring arbitrators to disclose AI use. Peru’s AI regulation, Decreto Supremo 115-2025-PCM (9 September 2025), does not mention arbitration, though it requires human oversight of high-risk AI decisions in sectors including justice (arts 28.11, 31.4), and the two Peruvian centres’ rules we read set no such duty. For India, see arbitration in India.^31,32,33

How does AI disclosure relate to an arbitrator’s duty of disclosure?

AI-use disclosure and the duty to disclose conflicts are separate. The conflict duty in the Arbitration Act 1996 (England, Wales and Northern Ireland) and the ICC, LCIA and KCAB rules covers circumstances that could raise doubts about impartiality (and, in the rules, independence: ICC art. 12(2), LCIA arts 5.4 and 5.5, KCAB art. 10.2). None of those provisions mentions AI.^11,16,20,34

Since 1 August 2025, s 23A has required an arbitrator, and anyone approached about an appointment, to disclose circumstances that “might reasonably give rise to justifiable doubts as to the individual’s impartiality”, in arbitrations begun from that date (s 23A(1) to (3); Arbitration Act 2025, s 17(4)). This is the arbitrator’s duty of disclosure; Halliburton v Chubb [2020] UKSC 48 had placed it within the statutory duty of fairness ([78], [81]).^{11,16,20,34,38,49}

Most AI texts ground disclosure in the right to be heard (SCC; SVAMC, p 20) and transparency (Ciarb para 7.2). In our reading, the English equivalents are s 33(1)(a), a reasonable opportunity to put one’s case, and s 37(1)(b), the chance to comment on a tribunal expert’s or adviser’s input.^1,2,7,35,36

Two texts we checked make impartiality a reason to disclose: the NAA, because a party could see a generative AI tool as “a source of outside influence” (Part II), and NCTDR and ICODR, for conflicts arising from AI algorithms or their developers (p 5). We found no ruling on s 23A and AI use; see AI in arbitrator selection and conflict checks and arbitration in London.^12,13,34

What happens if AI use is not disclosed?

Within the arbitration, the tribunal decides what follows a failure to disclose AI use. Where a party breaches a disclosure direction, Ciarb lets the tribunal inquire, take remedial steps, draw adverse inferences or reflect the failure in costs (paras 7.6, 7.7).²

We found no court decision setting aside an award because AI use was not disclosed. In ARIHQ c. Santé Québec, 2026 QCCS 1360, the Superior Court of Québec annulled an award whose authorities did not exist or did not support it, finding that the arbitrator’s authority had been delegated ([102] to [116]); the judgment does not say which tool, if any, was used. In LaPaglia v. Valve Corporation, a petition to vacate an AAA award alleging that the arbitrator had outsourced the decision to AI was dismissed on 9 December 2025 without a ruling on AI (awards challenged over AI use).^40,41

The nearest analogy we found is not an AI case: in DJP and others v DJO [2025] SGCA(I) 2, relying on material from parallel cases never put to the parties breached the fair hearing rule ([82]); see arbitration in Singapore. In our reading, a party could seek to set aside an award under s 68 of the 1996 Act, seek vacatur under 9 U.S.C. section 10(a)(4), or resist enforcement under art. V(1)(b) of the New York Convention where it was “unable to present his case”.^37,42,43,44

How should tribunals, parties and experts handle AI disclosure?

Settle disclosure in PO1 using the checklist above, then:

• Tribunals: say in advance how you may use AI, as ARB/24/46 does (s 28), and disclose before relying on AI-generated information outside the record (SVAMC Guideline 7). Where California’s section 1282.1(b)(2) applies, this is a statutory duty for generative AI from 1 January 2027.
• Parties and experts: note the tool’s name, version and settings, how it was used, and the prompts and outputs (SVAMC Guideline 3); parties should also check the output (Guideline 4).
• Labour-management arbitrators: disclose substantive generative AI use once planned, and get the parties’ consent before AI helps decide (NAA Advisory Opinion No. 28).
• Check for revised texts: Ciarb’s call for feedback on its Guideline ran to 26 May 2026, and the LCIA plans draft revised rules for comment later in 2026.^{1,12,19,23,28,29,48}

Not settled: what “appropriate disclosures” means under section 1282.1(b)(2), whether non-disclosure alone is enough to challenge an award, if prompts are privileged, and what revised Ciarb or LCIA texts will say.