CLA News / Eavesdropping Algorithms: AI’s Challenge to Mediation Ethics and Confidentiality By Joel Hames

Introduction

A recent piece of post-mediation commentary by mediator Julia Barnes raised a number of concerns about the use of AI in and around mediation.[1] The cornerstone of mediation as a method of alternative dispute resolution (ADR) lies in confidentiality. It is confidentiality that allows both participants and the mediator to have confidence in the process and in each other, thereby facilitating an environment conducive to resolution. AI presents a number of ethical and practical issues for mediation, and this piece seeks to identify and engage with those concerns.

This article will begin by looking at why confidentiality is integral to the process of mediation, how it protects those involved and how it contributes to the effective resolution of even the most entrenched disputes. Then moving to the commentary of Julia Barnes, as well as the comments of other practitioners surrounding their experiences and concerns with AI in mediation will look at how AI is being used in the mediation process. Additionally, this piece will examine the ethical concerns the AI usage generates, as well as the obligations and duties upon the mediator when such instances arise.

The Confidentiality Cornerstone

Confidentiality is one of the central features of mediation and perhaps one of the principal reasons why the process is so effective at resolving disputes. It encourages candor, allows parties to explore settlement positions without committing themselves irreversibly, and distinguishes mediation from the public and adversarial nature of litigation. Save for safeguarding concerns and illegality, confidentiality in mediation is absolute, or at least should be. The preservation of confidentiality remains essential to the legitimacy and practical utility of mediation as a form of dispute resolution.

The origins of confidentiality in mediation lie in the nature and development of mediation itself. Mediation emerged as a consensual process designed to facilitate negotiated settlement rather than impose adjudicated outcomes.[2] Given the process depends upon voluntary participation, (although the voluntary nature of mediation is shifting given many jurisdictions willingness to mandate court annexed mediation)[3] and the willingness of parties to reveal interests, concerns, and possible concessions, confidentiality became an organic part of its design.[4] Unlike a full trial, mediation seeks to create a protected forum in which parties may speak more openly than would be possible in ordinary adversarial proceedings. That protection has long been seen as necessary to encourage meaningful dialogue and to maximize the prospect of settlement.

This is embodied in the ‘without-prejudice rule’ that shields all mediations. The without-prejudice principle protects the statements, offers and/or other negotiated arrangements being subsequently adduced in evidence against the party who made them if negotiations fail.[5] Both the need and utility for such a provision is self-evident, parties are being increasingly encouraged to settle disputes rather than litigate them to judgment, and they will only earnestly negotiate if they are not later exposed to a material disadvantage should the mediation be unsuccessful.[6] In mediation, the ‘without prejudice’ principle performs a particularly important role because many discussions involve tactical concessions, admissions made for settlement purposes, or proposals advanced purely to test whether a compromise can be reached.[7] While confidentiality and without-prejudice privilege are not identical, they overlap significantly in practice and together provide the legal basis and normative shelter that allows mediation to function.

Confidentiality brings myriad benefits, and maintaining it throughout the process is essential. Primarily, confidentiality promotes honesty. A party is far more likely to disclose commercial pressures, personal motivations, or evidential weaknesses if assured that these matters will not later be deployed in open court. Secondly, confidentiality can protect relationships and reputation. In family, employment, commercial, and community disputes, preserving privacy may reduce hostility and make future cooperation possible. It can maintain the public perception and reputational integrity of parties that may be put in peril in a full public trial. Additionally, confidentiality encourages creativity. Mediated settlements often succeed because they move beyond the binary outcomes available through adjudication, and such creativity requires a secure setting in which proposals can be explored, revised, and abandoned without reputational or legal cost.

By fostering a sense of trust between conflicting parties and cultivating an environment amenable to resolution, the likelihood that parties will engage seriously with the process increases and thereby supports and upholds the wider objective of efficient and effective dispute resolution. If parties believed that mediators could routinely be called to give evidence, or that discussions in caucus or joint session might later be disclosed, mediation would quickly lose much of its distinctive value and be rendered ineffectual.

AI as Digital Eavesdropper

In light of the importance of confidentiality, recent comments by Julia Barnes raise practical and ethical questions about the future of mediation. Recalling a recent mediation she notes, ‘I had a moment in a recent mediation… There was another voice in the room, and they hadn’t signed the confidentiality agreement. It wasn’t a friend or family member – it was AI.’[8] Barnes goes on to explain how the confidentiality concerns she identified materially manifested, ‘I realized one of the participants had been feeding parts of the discussion into AI. When I returned to continue the discussion in their room, the mediation participant had clearly been using it to generate prepared responses.’[9] She objected to this clandestine participant stating, ‘Mediation depends on everybody being fully present in the flow of the conversation. It requires everyone to listen to multiple nuances and layers of conversation. Those discussions sometimes dart about, and they can be complex. When responses are shaped elsewhere by a machine, something fundamental can be lost.’[10]

Barnes further notes the impact of this intrusion upon her acting as the mediator and on the capacity to effectively mediate the dispute in a manner in which she preferred and/or is accustomed. She writes ‘I felt that I couldn’t really get an effective conversation going. We would discuss a point (for example the value of the property for negotiation purposes), make progress on it and I would feel that we were moving on, only for the same point to come back up again later. It felt like it had never been discussed and acknowledged. I was becoming increasingly perplexed… Towards the very end of the mediation, I realized what was happening, but I only had a chance to process it after the mediation had finished.’[11]

Barnes’ recent comments seek to raise confidentiality as an area of possible concern moving forward; however, she does not dismiss the use of AI out of hand. ‘I don’t think the answer is to ban the use of AI in all mediations. There are skilled mediators encouraging AI into the process. Susan Guthrie actively and openly uses AI in her matrimonial mediations.[12] She reports that by inviting AI into the room, the parties switch from blaming each other to shared problem solving mode.’[13] Instead, Barnes acknowledges to the seemingly inevitable rise of AI, noting, ‘At one level, this is unsurprising. AI is now everywhere. Of course it has found its way into mediation rooms.’ She is careful to highlight the possible uses of AI when she says ‘Used thoughtfully, AI could have a place. It might help a client organize their thoughts, reflect during breaks, or find words for something they are struggling to express. It may generate options for discussion which may help cut through blind spots leading to tunnel vision.’[14] Barnes concludes her comments by posing the question: ‘If discussions are being fed into AI tools, where is that information going? Who else is “in the room”?… So how can we use it [AI] in a way that supports the mediation without compromising confidentiality or trust?’[15]

Ethical Duties/Concerns for Mediators

Professional bodies and regulators have become concerned by the developments outlined above. Recent guidance from the International Bar Association (IBA) is notable as it purposely fails to endorse or reject AI outright, instead it frames AI use as permissible only where it remains subordinate to mediation’s core values of neutrality, confidentiality, informed participation, and human control.[16] The IBA guidelines recommend that users review privacy rules and terms before using AI in support of mediation, and make clear that AI safeguards are complementary to, rather than a substitute for, existing legal and ethical obligations.[17] Mediation agreements typically bind the parties, their representatives and, sometimes, experts or observers to confidentiality, but they are premised upon human participants capable of giving consent. Current AI programmers, in whatever form or variety, cannot sign such agreements, however if deployed by one or all of the participants they record, recall and store all the session data, including sensitive disclosures made under the protective without-prejudice umbrella.[18]

The concern gains traction when considering how large language models (LLMs) are trained and operated. User inputs to tools like ChatGPT or similar platforms are often retained, potentially incorporated into model fine-tuning/subsequent training or accessible to vendors under terms of service.[19] Even if not actively stored, confidential material risks regurgitation through memorization, where models reproduce training fragments in responses (a phenomenon documented in privacy audits of LLMs).[20] This undermines mediation’s absolute confidentiality, as disclosures intended to remain private may persist indefinitely in opaque systems, vulnerable to breaches, subpoenas, or adversarial extraction. Experts note that AI cannot distinguish confidential from non-confidential inputs, amplifying the risk in high-stakes disputes.[21] In these circumstances, AI can operate as an ‘eavesdropper’, akin to an uninvited third party, raising difficult questions as to whether its involvement breaches confidentiality obligations or affects any applicable privilege.

An additional concern is competence, that of the mediator and the participants. Mediators and representatives who use AI without understanding its limits may breach the standard of skill and care expected of them. The danger is not only that they may rely on inaccurate outputs, but that they may fail to appreciate data protection risks, vendor terms, or the ethical significance of outsourcing parts of the process to a machine. The UK government’s guidance for regulators on AI underscores the need to develop and implement proportionate, sector-specific oversight.[22]

It falls therefore to professional bodies/accreditation providers to provide a usage framework for existing members and practitioners, as well as update accreditation courses and CDP training to reflect and engage with AI and, if deemed appropriate, its correct usage. A number of associations have already taken steps in this direction, for example the American Arbitration Association (AAA) and International Mediation Institute (IMI) have published comprehensive standards for the responsible use of AI.[23] Further, the discussions within forums such as the Colloquium on the Use of Artificial Intelligence (AI) in Dispute Resolution and Remote Hearings in Arbitration and Mediation, convened by UNCITRAL demonstrate not only the growing practical application of AI in dispute resolution but also the institutional recognition of its significance.[24] By formally addressing procedural guidance and best practices for AI integration, states and international bodies are signaling that AI in mediation is no longer a theoretical discussion; it is a pressing, policy-level priority.[25]

Conclusion

AI’s unauthorized or undisclosed presence in mediation can undermine the core tenet of confidentiality, central and crucial to the mediation process. It should be stressed that AI is a tool, and it is best understood as such. One that clearly has its use cases and place within mediation. However, this should be at the discretion of the participants and the mediator as to whether they feel this appropriate. As a tool AI is incapable of signing the confidentiality agreement, yet fed sensitive disclosures becomes an uninvited third party, its opaque operations threatening the without-prejudice shelter essential to candor and possible settlement. Julia Barnes rightly asks, “Who else is in the room?”. Absent express authorization and ironclad safeguards, AI ceases to assist and begins to erode trust itself. It is important that every effort is made to integrate AI in a way that supports mediation without compromising confidentiality or trust.

Joel Hames

Joel Hames (Umesnited Kingdom) is a Bar Course Student at the Institute of Professional Legal Studies, Queen’s University Belfast. He is also an accredited mediator, member of the Society of Mediators and an International Associate of the Association of Arbitrators (Southern Africa) – jhames01@qub.ac.uk

[1] Julia Barnes, (LinkedIn, April 2026) <https://www.linkedin.com/feed/update/urn:li:activity:7447176341852147712/> accessed 2 May 2026.

[2] Laurence Boulle and Miryana Nesic, Mediation: Principles, Process, Practice (Lexis Publishing 2000).

[3] Churchill v Merthyr Tydfil Borough Council [2023] EWCA Civ 1416

  Directive Introducing Mandatory Mediation in the Gauteng Division of the High Court [2025] (Gauteng    Division, South Africa) 22 April 2025.

[4] (n 2) Boulle

[5] ibid

[6] ibid

[7] Ibid

[8] (n 1) Barnes

[9] ibid

[10] ibid

[11] ibid

[12] Akshatha Achar, ‘Practical Applications of AI in Mediation: CPR Committee Meeting Highlights’ (CPR ADR International Institute for Conflict Prevention & Resolution, 28 June 2024) <https://www.cpradr.org/news/practical-applications-of-ai-in-mediation-cpr-committee-meeting-highlights> accessed 2 May 2026.

[13] (n 1) Barnes

[14] ibid

[15] ibid

[16] International Bar Association Mediation Committee, ‘Guidelines on the Use of Generative Artificial Intelligence in Mediation’ (International Bar Association, 18 June 2025) <https://www.ibanet.org/Guidelines-on-the-use-of-generative-artificial-intelligence-in-mediation> accessed 2 May 2026.

[17] ibid

[18] Michelle Tong, ‘AI Tools and Privilege in the UK – What Are the Risks?’ (Osborne Clarke, 12 March 2026) <https://www.osborneclarke.com/insights/ai-tools-and-privilege-uk-what-are-risks> accessed 2 May 2026.

[19] ibid

[20] ibid

[21] ibid

[22] Department for Science, Innovation and Technology, ‘Implementing the UK’s AI Regulatory Principles: Initial Guidance for Regulators’ (Policy Paper, 2024) https://www.gov.uk/government/publications/implementing-the-uks-ai-regulatory-principles-initial-guidance-for-regulators accessed 2 May 2026.

[23] American Arbitration Association, ‘AAAi Standards for AI in ADR’ (2024) https://www.adr.org/media/sx2mjcdj/aaai_standards_for_ai_in_adr_.pdf accessed 2 May 2026.

[24] United Nations, ‘UNCITRAL Colloquium on the Use of Artificial Intelligence (AI) in Dispute Resolution and Remote Hearings in Arbitration and Mediation’ (United Nations Commission on International Trade Law) <https://uncitral.un.org/en/colloquium_aidisputeresolution> accessed 2 May 2026.

[25] Victoria Peña Morante, ‘The Rise of AI in Mediation: Friend or Foe?’ (International Mediation Institute, 30 April 2026) <https://imimediation.org/2026/04/30/the-rise-of-ai-in-mediation-friend-or-foe/> accessed 2 May 2026.