AI in Danish Dispute Resolution (2026): Evidence, Confidentiality and Admissibility

AI in dispute resolution Denmark has moved from a theoretical concern to a live, operational risk for companies and their advisers in 2026. As generative AI tools embed themselves in corporate workflows, from drafting correspondence to summarising contracts and generating analytical outputs, the material they produce increasingly ends up in the evidentiary record of commercial disputes. This guide gives practitioners and business leaders a grounded, Denmark-specific view of how AI-generated material is treated as evidence, how confidentiality is managed differently in courts and arbitration, and what concrete steps to take before, during and after a dispute.

For readers new to the field, a dispute resolution lawyer is a specialist who resolves commercial conflicts through litigation, arbitration, mediation or other alternative dispute resolution (ADR) mechanisms, advising on strategy, procedure, evidence and enforcement.

The four principal types of dispute resolution are litigation (proceedings before the ordinary courts), arbitration (binding private adjudication by a tribunal), mediation (facilitated settlement), and expert determination (a technical decision by an appointed specialist). The treatment of AI-origin evidence varies meaningfully across these forums, and that variation sits at the heart of this guide.

Executive summary: key risks and action steps

When AI-generated or AI-processed material becomes relevant to a dispute, the window for effective action is narrow. The following immediate steps protect evidentiary value and limit downstream exposure.

• Preserve immediately. Capture full system images, logs, prompts, outputs and underlying datasets before anything is overwritten or auto-deleted. Suspend routine retention and deletion routines for relevant systems.
• Document the AI tools involved. Record which models, versions, vendors and configurations generated or processed the material, and when.
• Engage experts early. Retain forensic and AI specialists to secure data and advise on reliability and authenticity before positions harden.
• Review and update contracts. Check dispute clauses, data-transfer terms, forensic-access rights and indemnities in key agreements.
• Control counsel use of AI. Restrict input of confidential client data into public large language models and route AI use through secure, approved environments.
• Consider protective measures. In arbitration especially, plan for confidentiality undertakings, redaction and protective orders over sensitive model data or trade secrets.
• Notify internal stakeholders. Alert legal, compliance and IT functions so preservation and governance decisions are coordinated and logged.

When to act: preservation steps should begin the moment a dispute is reasonably anticipated; expert engagement should follow promptly; and contractual and policy reviews should be underway shortly thereafter. Delay is the single most common cause of lost or degraded AI-origin evidence.

How generative AI and algorithmic tools create evidentiary issues

Understanding the evidentiary problems posed by AI in dispute resolution Denmark requires a short, non-technical grounding in how these systems behave. Generative AI systems predict plausible output based on statistical patterns in training data and on the prompt they receive. They do not retrieve verified facts, and the same prompt can produce different outputs at different times. That non-deterministic quality is precisely what makes AI-origin material contentious when it is offered as evidence.

What counts as “AI-generated” versus “AI-assisted” content

The distinction matters for both reliability and disclosure. AI-generated content is produced substantially by a model, a drafted narrative, a synthetic summary, a generated image or a computed analysis. AI-assisted content is human-authored material refined or edited with AI support, such as a contract reviewed by a tool before a lawyer finalised it. The evidentiary weight and the chain of accountability differ: AI-generated output raises sharper questions about who stands behind the content, while AI-assisted work product keeps a human author in the loop but may still carry algorithmic influence that a court or tribunal will want to understand.

Typical forensic indicators and metadata

Electronic evidence in Denmark is assessed in part through its technical signatures. For AI-origin material, the relevant signals include prompt logs, model and version identifiers, timestamps, API call records, system logs, document metadata and audit trails. These artefacts help establish when material was created, by which tool, and whether it was subsequently altered. Where metadata is absent or inconsistent, the party relying on the material faces a harder task in establishing authenticity and reliability.

Common failure modes

Three failure modes recur. First, hallucination, the model generates confident but fabricated content, including invented citations or facts. Second, unreliable attribution, output cannot be traced back reliably to a verifiable source or to a specific input. Third, drift and non-reproducibility, the same prompt yields materially different outputs over time, making it difficult to demonstrate that a given output was in fact produced as claimed. Each failure mode is a potential line of challenge for an opposing party and a potential weakness for the party relying on the material.

Admissibility of AI-generated evidence in Danish courts: legal framework

Danish civil procedure does not contain a bespoke code for artificial intelligence. Instead, AI-origin material is assessed under the general principles governing documentary and electronic evidence. The question practitioners face is not whether a special “AI rule” applies, but how established standards of authenticity, reliability and probative value are applied to material whose provenance is algorithmic.

Governing Danish rules

Civil litigation in Denmark is governed by the Administration of Justice Act (Retsplejeloven), the primary procedural statute available through Retsinformation. Danish procedure operates on a principle of free assessment of evidence: the court weighs the evidence before it and determines what probative value to attach, rather than applying rigid exclusionary rules of the kind found in some common-law systems. In practice this means AI-generated material is rarely excluded outright on categorical grounds; instead, questions about its reliability are typically reflected in the weight the court assigns to it.

Authenticity, reliability and probative value

For AI evidence admissibility in Denmark, three linked questions dominate. Authenticity: is the material what it purports to be, and can its origin and integrity be demonstrated through metadata, logs and chain of custody? Reliability: can the party show that the generating process produced a dependable output, or at least explain its limitations candidly? Probative value: what does the material actually prove, and how much should it move the court given its algorithmic origin? Because Danish courts assess evidence freely, a party that cannot establish authenticity may still have the material considered, but with substantially reduced weight, or the court may call for forensic analysis before relying on it.

Nordic and European authorities

There is, at the time of writing, no settled body of Danish case law specifically addressing generative AI output as evidence; this remains an evolving area, and counsel should treat it as advisory and seek current local advice. Danish courts can be expected to extend existing electronic-evidence reasoning to AI material, informed by the broader European regulatory direction. The European Commission’s European approach to artificial intelligence, including the EU AI Act, sets the cross-border regulatory context that increasingly shapes how AI systems are classified, documented and governed, factors that feed directly into how reliably their outputs can be demonstrated.

The Court of Justice of the European Union, through its jurisprudence on evidence and data protection, provides persuasive framing on balancing disclosure against privacy and public-interest considerations. It is reasonable to expect Danish courts to draw on documentation and transparency obligations emerging from the EU framework when assessing whether AI-origin evidence is reliable enough to carry weight.

AI evidence in arbitration: practical differences and confidentiality risks

Arbitration is where much high-value commercial dispute activity involving AI material will be resolved, and the handling of AI arbitration in Denmark differs from court litigation in several practically important respects. The most significant are confidentiality, the tribunal’s procedural flexibility, and the centrality of party-appointed experts. Danish arbitration is governed by the Danish Arbitration Act, and institutional disputes are frequently administered by the Danish Institute of Arbitration (Voldgiftsinstituttet).

Arbitration rules and the confidentiality default

Where Danish courts operate as public proceedings by default, arbitration is generally treated as confidential in practice, though the precise scope of confidentiality depends on the seat, the applicable institutional rules and any express agreement between the parties. This confidentiality is a core reason parties concerned about exposing proprietary AI systems, prompts or model data often prefer arbitration. It is not absolute: confidentiality may yield to public-interest obligations, regulatory requirements or enforcement proceedings. Parties should not assume that choosing arbitration automatically shields sensitive AI material; the protection must be actively secured through the procedural order and confidentiality undertakings.

How arbitrators assess electronic and AI evidence

Arbitral tribunals typically enjoy broad discretion over admissibility and tend to take a pragmatic view: rather than excluding AI-origin material, they admit it and determine weight, often guided heavily by party-appointed expert testimony. This makes the quality of forensic and technical evidence decisive. A party offering AI-generated material should be ready to explain, through a credible expert, how the output was produced, what safeguards existed, and why it should be trusted. A party challenging such material should deploy its own expert to expose hallucination risk, non-reproducibility or gaps in the audit trail.

Managing confidentiality and third-party AI vendor data

AI material frequently sits with third-party vendors, cloud model providers, analytics platforms or managed AI services. This complicates confidentiality and production because the data may be controlled, logged or stored outside the disputing party’s direct possession. Counsel should map vendor relationships early, confirm contractual rights of access and forensic capture, and plan for protective orders that extend confidentiality obligations to experts and vendor personnel who handle the material. The interaction between AI confidentiality in arbitration and third-party data control is one of the most under-appreciated risks in current practice.

As noted in the primer above, the four types of dispute resolution, litigation, arbitration, mediation and expert determination, each handle AI material differently; arbitration’s confidentiality and procedural flexibility often make it the preferred forum where sensitive AI systems are in issue.

Data protection, GDPR and Datatilsynet considerations

The intersection of data protection and disclosure is unavoidable whenever AI outputs contain personal data. The General Data Protection Regulation, together with the Danish Data Protection Act (Databeskyttelsesloven), applies to the processing of personal data in the context of litigation and arbitration, and the Danish Data Protection Agency (Datatilsynet) is the supervisory authority whose guidance counsel should consult.

Personal data in AI outputs and the legal basis for disclosure

AI-generated material routinely contains personal data, names, identifiers, or inferences about individuals, and producing it in a dispute is itself an act of processing that requires a lawful basis under the GDPR. There is no absolute bar to disclosure: the establishment, exercise or defence of legal claims is a recognised ground for processing. But the party disclosing must still satisfy proportionality, data minimisation and transparency obligations. The practical consequence is that AI material should not be produced wholesale; it should be scoped to what is relevant and necessary for the dispute.

Anonymisation versus pseudonymisation

Where personal data in AI outputs is not itself probative, counsel should consider removing or masking it. True anonymisation, rendering individuals no longer identifiable by any reasonable means, takes the data outside the GDPR’s scope. Pseudonymisation, replacing identifiers with keys that can be reversed, reduces risk but keeps the data within scope because re-identification remains possible. For disclosure in Danish disputes, understanding which technique has been applied is essential, because it determines the residual compliance obligations and the protective measures required.

Datatilsynet guidance and cross-border transfer traps

Datatilsynet publishes guidance on the use of AI and on the processing of personal data, which practitioners should treat as the authoritative Danish supervisory position. A recurring trap arises where AI material, or the vendor infrastructure holding it, sits outside the European Economic Area. Transferring such data for the purposes of a dispute may constitute a restricted international transfer requiring appropriate safeguards. Counsel should identify at the outset where model data and logs are physically stored and processed, and build any cross-border transfer compliance into the preservation and production plan rather than discovering the problem during disclosure.

Practical preservation, collection and production checklist for companies

The difference between a defensible and an indefensible evidentiary position in AI in dispute resolution Denmark is usually made in the first days after a dispute is anticipated. The following structured approach helps companies preserve, collect and produce AI-origin material credibly.

Immediate preservation steps

• Suspend automatic deletion and retention policies on systems holding relevant AI material.
• Preserve full system snapshots and complete logs, including prompt histories, API call records and audit trails.
• Where feasible, containerise or snapshot the relevant model configuration and the underlying datasets so the state at the relevant time is captured.
• Record the version, vendor and configuration of each AI tool involved.

Forensic capture and chain of custody

Preservation only protects value if it is done defensibly. Engage forensic specialists to capture material using sound methods, document who accessed what and when, and maintain an unbroken chain of custody. For electronic evidence in Denmark, the integrity of metadata and the demonstrability of the collection process directly affect the authenticity and weight a court or tribunal will attach to the material.

Producing prompts, models and training data

Parties should anticipate requests to produce prompts, outputs and, in some cases, model or training data. Such material can be commercially sensitive, so expect to negotiate scope aggressively and to rely on redaction, trade-secret protections and, in arbitration particularly, protective orders and in-camera review. Production should always be balanced against relevance, proportionality and the GDPR considerations discussed above.

Using technology-assisted review and producing work product

Large volumes of AI-related material make manual review impractical, and technology-assisted review in Denmark is increasingly central to efficient document production. TAR workflows, including predictive coding and continuous active learning, can accelerate review, but counsel must be able to defend the methodology if challenged. Document the review protocol, validation steps and quality controls so the process itself can withstand scrutiny.

The forum you are in materially changes how production and confidentiality operate. The table below summarises the key differences between Danish courts and arbitration.

Counsel use of generative AI: ethics, privilege and disclosure

The risks of generative AI in litigation in Denmark are not confined to the parties, they extend to how counsel themselves use AI. The Danish Bar and Law Society (Advokatsamfundet), together with the rules of conduct for lawyers (de advokatetiske regler) and the professional duties under the Administration of Justice Act, governs confidentiality and the handling of client data, and those rules apply squarely to the use of AI tools in legal practice.

When counsel’s use of AI risks waiver

Entering privileged or confidential client information into a public large language model may, depending on the platform’s data-handling practices, amount to disclosure to a third party. That can jeopardise confidentiality and, in a cross-border matter, potentially affect privilege. The safe working assumption is that anything typed into a public, consumer-facing AI tool may be retained and used by the provider, and should therefore never include confidential client material.

Best practices for secure use

• Use only secure, approved AI environments with contractual guarantees on data handling and non-retention.
• Never input identifiable client data or privileged material into public LLMs.
• Retain logs of AI use so the firm can demonstrate how tools were employed if later questioned.
• Verify AI output against primary sources before relying on it, given the risk of hallucinated citations and facts.

Drafting AI-use clauses for engagement letters

Firms and clients increasingly agree in advance how AI may be used in a matter. Engagement letters should set out the permitted scope of AI use, the environments allowed, confidentiality safeguards and record-keeping obligations. Clear terms protect both the client’s confidentiality interests and the firm’s position if the use of AI is later scrutinised.

How companies should prepare operationally: policy, contracts and evidence playbook

Preparation for AI in dispute resolution Denmark is an operational discipline, not a reaction to litigation. Companies that build the right governance, contractual terms and playbooks before a dispute arises will preserve evidence more effectively and face fewer compliance surprises.

Contract clause checklist

• Evidence preservation. Obligations on counterparties and vendors to preserve AI logs, prompts and outputs on notice of a dispute.
• Forensic access. Rights to access and forensically capture AI systems and data held by vendors or counterparties.
• Data transfers. Terms addressing the location of model data and the safeguards required for any cross-border transfer in a dispute.
• Indemnities. Allocation of liability for AI outputs, including where third-party tools produce erroneous or infringing material.

Internal governance and training

Companies should maintain an AI use policy covering approved tools, logging requirements, data-handling rules and preservation triggers. Staff who use AI in roles that may generate disputable material should be trained to understand that their prompts and outputs may one day be evidence. Governance should assign clear ownership for preservation decisions when a dispute is anticipated.

When to appoint an AI forensic expert

An AI forensic expert should be engaged as soon as AI-origin material appears likely to be central to a dispute, ideally before any collection takes place, so capture is defensible from the outset. Early expert involvement strengthens authenticity arguments, anticipates the opposing party’s challenges, and informs the reliability narrative that courts and tribunals will weigh.

Conclusion: recommended next steps for AI in dispute resolution Denmark

AI in dispute resolution Denmark is a fast-moving, largely unsettled area where sound preparation decisively outperforms reactive crisis management. Danish courts and arbitral tribunals will assess AI-origin material through established principles of authenticity, reliability and probative value, with arbitration offering greater confidentiality but demanding robust expert evidence, and with the GDPR and Datatilsynet guidance shaping every disclosure decision. For general counsel, three priorities stand out: first, implement preservation triggers and an AI governance policy now, so evidence is captured defensibly the moment a dispute is anticipated; second, review and update contractual dispute, data-transfer and forensic-access clauses in key agreements; and third, control counsel and employee use of AI through secure environments and clear engagement terms.

Companies that act on these before a dispute crystallises will be markedly better positioned to preserve, produce and, where necessary, challenge AI material.

This is general information and not legal advice. Given the unsettled state of Danish authority on AI-origin evidence, companies and counsel should seek specific local advice on their circumstances.

Sources

1. Retsinformation, Danish government legal information
2. Datatilsynet (Danish Data Protection Agency)
3. Domstol.dk (The Danish Courts)
4. Advokatsamfundet (The Danish Bar and Law Society)
5. Danish Institute of Arbitration (Voldgiftsinstituttet)
6. European Commission, European approach to artificial intelligence
7. Court of Justice of the European Union (Curia)
8. General Data Protection Regulation (Regulation (EU) 2016/679)