AI Ethics for Mediators: A Practical Guide to Responsible Use

Artificial intelligence is becoming part of the way people prepare for difficult conversations. Clients can arrive at mediation with an AI-generated position statement, a suggested settlement figure or a confident explanation of their legal rights. Mediators, meanwhile, may be exploring tools that promise quicker administration, clearer communication and help with preparation.

These developments create opportunities, but they also raise questions about trust, confidentiality and professional judgement. Understanding AI ethics does not require mediators to become technology specialists. It requires us to recognise where a tool can help, where it can cause harm, and what safeguards a responsible practice needs.

AI ethics for mediators - start with the principles of mediation

The starting point is familiar: impartiality, confidentiality, informed participation and the parties’ ability to make their own decisions. Any proposed use of AI should support these principles. If it weakens them, convenience is insufficient justification.

Generative AI produces responses from patterns in data. Fluent language does not establish that an answer is accurate, balanced or appropriate. A system can invent a legal authority, misunderstand a document or present an assumption as fact. The mediator remains responsible for anything they adopt, share or allow to shape their practice.

Consider the difference between asking AI to simplify a generic explanation of mediation and asking it to assess which participant is being unreasonable. The first may assist communication. The second risks introducing unsupported judgements into a process that depends on careful listening and impartiality.

When might a mediator use AI?

Begin with tasks that do not require confidential case information. AI can help structure a public information article, improve a generic appointment email, suggest questions for a fictional training scenario or produce an initial checklist for organising a mediation day. Each output still needs review for accuracy, tone and accessibility.

Work involving live cases requires a more demanding assessment. Summarising documents, translating statements or drafting an outcome record may appear administrative, but errors can change meaning or omit an important qualification. Use an appropriately assessed system, resolve confidentiality and data protection requirements, and check the result against the original material.

Avoid relying on AI to assess credibility, diagnose someone’s emotions or determine a fair settlement. An apparently neutral recommendation can anchor discussion around a figure or interpretation that has no sound basis. Parties must retain ownership of their decisions, with independent legal advice where appropriate.

Managing clients’ use of AI

Raise the subject during preparation. A neutral question such as, “Have you used any AI tools to help prepare, or are you planning to use them during the session?” opens a useful conversation without making participants defensive.

Clients may use AI because they feel overwhelmed, cannot afford extensive advice or need help expressing themselves. Acknowledge that purpose. Then explore what information they supplied, what the tool produced and whether important claims have been checked. Do not assume polished wording means the person fully understands or agrees with it.

Where a participant relies on an AI-generated prediction, ask what supports it: which jurisdiction, which facts, which sources and which assumptions? Treat the prediction as something to examine rather than established evidence. The Civil Mediation Council’s April 2026 consultation response similarly identifies disclosure of AI use as important when discussing predictions in mediation.[1]

Agree practical boundaries before the session. Address uploading documents, entering another person’s statements into a chatbot, using live AI assistance, and recording or transcription. Explain the confidentiality implications. A participant’s decision to use a tool does not authorise them to share everyone else’s information with its provider.

For example, a client may bring an AI-written statement describing their manager as manipulative. Rather than debating the label, invite them to identify the behaviour, explain its impact and describe what they need to change. This brings the discussion back to their experience. It also helps distinguish the client’s own concerns from interpretations introduced by a system that has heard only one account.

Data privacy requires more than removing names

Mediation files can contain health information, financial details, allegations and commercially sensitive material. Before entering anything into AI, establish what the provider receives, how it uses that information and whether the proposed processing is lawful and consistent with confidentiality obligations.

Check the relevant product and account terms. Is information used for model training? Who can access it? Where is it processed? How long is it retained? Can it be deleted? What contractual protections and international transfer arrangements apply? A paid subscription or a “training off” setting does not, by itself, answer every question.

Removing names may reduce risk without making information anonymous. A distinctive job title, location and sequence of events can still identify someone. Wherever possible, use fictional examples or genuinely anonymised material, and provide only the information necessary for the task.

Permission to use AI within the mediation process and a lawful basis for processing personal data are separate matters. Participant agreement does not automatically satisfy UK GDPR requirements. Identify the appropriate lawful basis; special category information, such as health data, also requires an additional condition. Assess whether a data protection impact assessment is required, particularly where processing is likely to create high risks for individuals.[2]

Keep recording and transcription under control

AI meeting assistants can join online sessions automatically. Their presence should never be treated as an ordinary technical detail. A transcript may capture private caucus discussions, settlement proposals and comments participants expected would remain within the conversation.

Check meeting settings and attendees, disable automatic assistants unless specifically agreed, and explain the position to everyone involved. Where transcription is justified, establish its purpose, access controls, retention and deletion arrangements beforehand. Consider whether a less intrusive accessibility adjustment would meet the need. If an unauthorised tool appears, pause, stop further capture and establish what was collected; assess any confidentiality or personal data breach and the response required.

Five practical habits for responsible use

1. Write a short AI policy. Explain permitted uses, prohibited uploads, approval arrangements and how concerns will be handled. Make it understandable to clients and anyone supporting your practice.

2. Test with fictional material first. Explore how the tool behaves before involving live work. Notice omissions, invented details and changes in meaning, including across languages or communication styles.

3. Verify before relying. Check facts, sources, calculations and summaries. Ask whether an output reflects incomplete information or reinforces one party’s account. Confident wording is no substitute for evidence.

4. Discuss fairness and choice. Consider whether AI use creates pressure or an advantage that affects meaningful participation. Explain proposed uses clearly, allow questions and provide an appropriate alternative where someone objects.

5. Review your arrangements regularly. Keep proportionate records of material uses and safeguards, check changing supplier terms, and revisit relevant professional guidance. Bring difficult ethical questions into supervision or peer discussion.

A suggested AI clause for your Agreement to Mediate

Clear written boundaries can help everyone understand what is permitted before mediation begins. The following model is a starting point for adaptation and legal review alongside your existing Agreement to Mediate and privacy notice. It is not a mandatory professional code or a substitute for assessing a particular tool.

Use of Artificial Intelligence and Automated Tools

1. Scope and responsibility

For this clause, “AI tools” includes generative AI services, automated meeting assistants, and tools that record, transcribe, summarise or analyse mediation communications. The Mediator and Parties remain responsible for their use of these tools and for complying with this Agreement.

2. Confidential information

Neither the Mediator nor any Party shall upload, enter or otherwise disclose confidential mediation information to an AI tool without the prior written agreement of the Mediator and all Parties, together with any other authorisation required. This includes documents, private discussions, offers, personal information and settlement terms. Removing names alone does not necessarily make information anonymous.

3. Recording and live assistance

AI recording, transcription, meeting assistants and live analysis shall not be used during any mediation meeting, including private meetings, unless specifically agreed in writing beforehand by the Mediator, all Parties and any other participant affected. Automatic recording and meeting assistants must otherwise be disabled.

4. Agreement to a proposed use

Before permission is sought, the person proposing the use shall explain the tool, purpose, information involved, material risks and proposed safeguards. Any written agreement shall specify who may access the information and outputs, applicable retention and deletion arrangements, and any restrictions on further use or sharing. Permission for one purpose does not authorise another use. Information from a private meeting remains subject to the separate confidentiality arrangements governing that meeting.

5. Preparation and AI-generated material

Parties may use AI to assist their preparation, subject to their confidentiality and data protection obligations. A Party presenting AI-generated legal authorities, factual assertions, calculations or outcome predictions shall identify that use and take reasonable steps to verify the material. AI outputs shall not be treated as authoritative merely because they were produced by an AI tool.

6. Human judgement and participation

The Mediator shall retain responsibility for conducting the mediation impartially and exercising independent professional judgement. AI shall not determine credibility, impose a settlement or replace the Parties’ own decisions. Participants may raise concerns about proposed AI use and request an appropriate alternative. Any participant affected may withdraw permission for future use, following which that use shall stop while alternative arrangements are considered.

7. Data protection and concerns

Agreement under this clause does not, by itself, establish compliance with data protection law or authorise disclosure of another person’s information. Anyone becoming aware of unauthorised AI use shall notify the Mediator promptly and stop that use. The Mediator may pause the mediation to assess the circumstances and agree appropriate safeguards, without limiting any legal reporting or remedial obligations.

8. Other participants

Each Party shall ensure that its representatives and anyone attending at its invitation are informed of these requirements and agree to comply with them before participating.

Putting the model into practice

Record any approved exception in a short written schedule naming the tool, purpose, information, access and deletion arrangements. Signing the Agreement to Mediate should not constitute blanket permission for AI use. Permission within the mediation process remains separate from the lawful basis and any additional conditions required for processing personal data.

Protect the human value of mediation

AI can help mediators work more efficiently and make information easier to understand. Its value depends on thoughtful selection, clear boundaries and accountable human oversight.

At DRA, we encourage mediators to approach these developments with curiosity and professional care. The essential question is whether a particular use helps people participate confidently while protecting the trust on which mediation depends. If you need help, please book a discovery call and let's talk through your needs.

Sources and further reading

Further reading from the Civil Mediation Council and the Information Commissioner’s Office. Guidance checked on 21 October 2026.

1. Civil Mediation Council response to the Civil Justice Council consultation, 14 April 2026, paragraphs 14–16.

2. ICO: AI and data protection, including lawfulness. The ICO notes that its AI guidance is under review following the Data (Use and Access) Act; check current guidance before implementing a process.

3. CMC: Where is AI Heading?, June 2026: practitioner discussion of client AI use, confidentiality and positional thinking.