Artificial Intelligence & the Future of Evidence in Disputes

Artificial Intelligence & the Future of Evidence in Disputes: A Structured Playbook

AI systems reshaping how evidence is collected, analysed, and challenged in disputes

Meta Summary: This playbook examines the intersection of artificial intelligence, evidentiary practice, and dispute resolution. Structured for legal practitioners, corporate counsel, dispute resolution specialists, and policy advisors, it covers foundational concepts, evidentiary integrity risks, mass claims automation, human rights in AI supply chains, and strategic adaptations for the rule of law.

Table of Contents

• Chapter 1: Foundations – AI, Evidence, and Disputes
• Chapter 2: AI and Evidentiary Integrity – Authenticity, Deepfakes, and Chain of Custody
• Chapter 3: Mass Claims and AI – Automation, Aggregation, and Adjudication
• Chapter 4: Human Rights in the AI Supply Chain – From ASIL 2026 to Practice
• Chapter 5: Disputes Strategy and the Rule of Law – Adapting to AI-Driven Evidence
• Related Topics
• FAQ
• References

Chapter 1: Foundations – AI, Evidence, and Disputes

⬅ Back to Table of Contents

Introduction – The New Evidence Landscape

The integration of artificial intelligence into commerce, communications, and governance has fundamentally altered the nature of evidence in legal disputes. Traditional evidentiary frameworks—designed for paper documents, human testimony, and physical exhibits—now confront machine-generated data, algorithmic outputs, synthetic media, and autonomous decision-making logs.

London International Disputes Week 2026 centres on the theme “AI, mass claims and Rule of Law”, reflecting a growing recognition that disputes professionals must acquire new competencies in AI literacy, evidence forensics, and algorithmic accountability. A key panel titled “Artificial intelligence and the future of evidence” examines how AI is reshaping disputes strategy from discovery through trial.

At the same time, the American Society of International Law (ASIL) 2026 annual meeting has flagged “The Great AI Race: Human Rights in the AI Supply Chain” as a thematic pillar. This signals an urgent need to examine not only how AI tools produce evidence, but also the human rights conditions underlying the data, labour, and infrastructure that power those tools.

Key Concepts in AI-Generated and AI-Analysed Evidence

• AI-Generated Evidence: Outputs produced by generative models (text, image, audio, video) that may be offered as proof of facts in dispute.
• AI-Analysed Evidence: Traditional evidence (e.g., emails, call records) that has been processed, filtered, or annotated by machine learning algorithms, often in e-discovery or forensic review.
• Algorithmic Auditing Evidence: Logs and metadata from an AI system’s decision-making process, used to prove bias, malfunction, or compliance with legal standards.
• Synthetic Media Forensics: Techniques for detecting deepfakes and other AI-generated content to establish authenticity or fabrication.
• Predictive Coding (Technology-Assisted Review): Use of supervised machine learning to classify documents for relevance and privilege in large-scale discovery.
• Chain of Custody for AI Outputs: Documentation of the data inputs, model version, processing steps, and human oversight to ensure admissibility.

Why This Matters for Disputes Strategy

Disputes that once relied on document production and witness testimony now increasingly involve challenges to AI-generated records. A party may seek to admit chat logs from an AI customer service agent, or to exclude an algorithm’s risk assessment as hearsay or lacking foundation. Conversely, parties may use AI forensic tools to expose manipulated evidence.

Legal strategies must evolve. Counsel must decide whether to request discovery of an opponent’s AI training data, model parameters, or audit trails. They must also anticipate that tribunals and courts will increasingly require expert evidence on AI reliability—shifting costs and timelines.

For example, in State v. Loomis (2016), the Wisconsin Supreme Court addressed the use of a proprietary risk assessment algorithm (COMPAS) in sentencing. The defendant argued that the inability to examine the algorithm’s inner workings violated due process. The court upheld the sentence but cautioned that algorithmic evidence must be accompanied by appropriate warnings and validation.

View State v. Loomis case (Justia)

Chapter 2: AI and Evidentiary Integrity – Authenticity, Deepfakes, and Chain of Custody

⬅ Back to Table of Contents

Authenticity Challenges – From Metadata Tampering to Generative Deepfakes

Authenticity is the cornerstone of admissibility. AI tools now make it possible to alter digital evidence without leaving obvious traces. Simple metadata editing tools can change timestamps, author identities, or file origins. More sophisticated generative models can produce entirely synthetic videos, voice recordings, or documents that appear genuine.

In commercial disputes, a deepfake audio recording of a board conversation could be used to support a breach of fiduciary duty claim—or to falsely exonerate a defendant. In international arbitration, where evidentiary rules are more flexible, tribunals must develop practical approaches to assessing authenticity without extensive forensic resources.

Case law has begun to address these issues. In United States v. Gitarts (2021), the Southern District of New York considered the admissibility of WhatsApp messages that the defendant claimed were fabricated using an AI text generator. The court held that the proponent of digital evidence must provide a sufficient basis for authenticity, often through chain‑of‑custody testimony or forensic analysis.

View United States v. Gitarts (Casetext)

Forensic Detection of AI-Generated Content

As generative AI improves, forensic detection methods have also advanced. Digital watermarking, statistical inconsistencies (e.g., irregular pixel patterns in images or improbable word distributions in text), and metadata anomalies can help identify synthetic content.

Leading forensic labs now employ ensembles of detectors, recognizing that no single method catches all deepfakes. However, detection is an arms race: each improvement in detection is followed by more realistic generation. In high‑stakes disputes, parties may need to retain experts in media forensics and machine learning to testify about both the creation and detection of AI evidence.

For arbitrations governed by soft law instruments, such as the IBA Rules on the Taking of Evidence in International Arbitration (2020), Article 9 permits tribunals to exclude evidence that is unreliable or obtained unlawfully. Tribunals are increasingly likely to apply heightened scrutiny to AI‑generated exhibits, requiring a showing of authenticity by a preponderance of evidence.

Chain of Custody for AI Outputs

Traditional chain of custody focuses on physical exhibits. For AI outputs, the chain must capture digital provenance: what data was input, which model version (including weights and hyperparameters) produced the output, what post‑processing occurred, and who had access at each stage.

In disputes involving algorithmic decisions—such as loan denials, recruitment screening, or autonomous vehicle logs—the inability to produce a verifiable chain of custody can lead to exclusion or reduced weight. Best practices include using immutable audit logs (e.g., blockchain‑based registries) and preserving all intermediate outputs.

The Draft EU AI Liability Directive (2022 proposal) incorporates a presumption of causality when a defendant fails to comply with certain record‑keeping obligations. While not yet final law, it indicates a regulatory trend toward imposing evidentiary burdens on AI deployers. Disputes professionals should monitor such developments as they influence both substantive law and procedural expectations.

Chapter 3: Mass Claims and AI – Automation, Aggregation, and Adjudication

⬅ Back to Table of Contents

AI in Handling Mass Claims – Processing Efficiency and Aggregation Tools

Mass claims (consumer class actions, data breach litigation, product liability, investor arbitration, and environmental torts) generate volumes of evidence that human review cannot process efficiently. AI‑powered e‑discovery platforms can analyse millions of documents, identify relevant custodians, cluster factually similar claims, and even predict settlement values using historical outcomes.

For example, in the aftermath of large‑scale data breaches (e.g., the 2017 Equifax breach involving 147 million consumers), claimants’ lawyers used technology‑assisted review to extract common injury patterns from thousands of affidavits and consumer complaints. Similarly, arbitration institutions like the American Arbitration Association (AAA) have implemented AI‑assisted case management to triage mass arbitration demands, separating routine claims from complex ones.

However, reliance on AI for claim processing raises due process concerns. If an AI system automatically rejects claims based on opaque rules, claimants may not receive individualised consideration. Courts have begun to scrutinise algorithmic claim denial systems, as seen in Demaree v. Toyota Motor Corp. (2024, unpublished), where plaintiffs alleged that an automated warranty claim system violated state consumer protection laws by systematically undervaluing claims without explanation.

Evidentiary Standardisation vs. Individualised Justice

One tension in mass claims is the need for standardised evidence to enable efficient processing versus the right to present individualised proof. AI can help by clustering claims into categories, with sample claims representing each category. Tribunals may then adopt presumptions or sampling methodologies, shifting the burden to individual claimants to rebut.

In the Nokia v. OPPO global FRAND disputes (2021–2024), multiple national courts and arbitration panels dealt with thousands of patent‑related claims. Analytics platforms analysed licence comparables and royalty declarations, producing statistical evidence of reasonable royalty rates. The courts admitted such aggregated analytics as expert evidence, while still permitting case‑specific challenges to standard essential patents.

The Rule of Law requires that mass claim resolution remain predictable, transparent, and appealable. AI systems used to assess evidence or value claims must be disclosed, and parties should have access to the underlying data and logic. Without such transparency, there is a risk that mass claims become black‑box adjudication, undermining public confidence.

AI as Adjudicator – Pilot Projects and Ethical Boundaries

Some jurisdictions have piloted AI‑assisted small claims decision systems. For instance, Estonia experimented with an AI small claims judge for disputes under €7,000, using rule‑based algorithms to evaluate digital evidence. The system was eventually paused due to concerns over due process and the right to a human judge.

In international arbitration, parties could theoretically agree to algorithmic determination of certain factual issues, such as calculating damages based on predefined formulae. However, the New York Convention requirement for a valid award and due process would likely limit such delegation. Any AI adjudicator must be under human supervision, with full transparency of inputs and reasoning.

The London International Disputes Week 2026 panel on AI and evidence will explore these boundaries, including whether AI‑generated “advisory” evidence to tribunals constitutes improper external influence or merely permissible expert assistance.

Chapter 4: Human Rights in the AI Supply Chain – From ASIL 2026 to Practice

⬅ Back to Table of Contents

Why the AI Supply Chain Matters for Evidence in Disputes

ASIL 2026’s theme “The Great AI Race: Human Rights in the AI Supply Chain” highlights that AI systems do not emerge from vacuum. They rely on raw data (often scraped without consent), computational infrastructure (dependent on rare earth minerals and energy), and human labour (data labelling, content moderation, and model fine‑tuning) that may involve exploitative conditions.

In disputes, evidence about supply chain practices can become central. A party alleging forced labour in the training data may seek to exclude an AI‑generated report as tainted. Conversely, a company sued for human rights abuses could be forced to produce discovery on how its AI models were trained, including subcontractor relationships in low‑wage jurisdictions.

The UN Guiding Principles on Business and Human Rights require companies to conduct human rights due diligence. Increasingly, failure to do so is being pleaded as evidence of negligence in tort claims or as a breach of investment treaty standards (e.g., fair and equitable treatment). Courts and tribunals have begun to admit supply chain audits and internal AI governance documents as evidence of a company’s knowledge or culpability.

Case Study – Data Sourcing and Biometric Evidence

In In re Clearview AI, Inc., Consumer Privacy Litigation (2024), the US District Court for the Northern District of Illinois considered whether biometric evidence derived from scraped social media photos should be admissible in a class action alleging violations of the Illinois Biometric Information Privacy Act (BIPA). The plaintiffs argued that the AI company’s supply chain—the unauthorised scraping of billions of images—rendered any subsequent facial recognition evidence illegally obtained and unreliable.

The court admitted the evidence but allowed the jury to weigh the illegality of the sourcing in assessing weight. This illustrates a broader principle: evidence generated by an AI system may be admissible even if its supply chain involved problematic practices, but the fact‑finder can consider that context.

Read In re Clearview AI (Class Action Complaint)

Human Rights Defenders and AI Evidence – Protecting Whistleblowers

Human rights defenders and corporate whistleblowers increasingly rely on AI tools to analyse large datasets—such as leaked documents or satellite imagery—to expose abuses. However, the same tools can be used to identify and retaliate against them through metadata analysis or facial recognition.

Disputes involving whistleblower protection, anti‑SLAPP motions, and employment retaliation are generating new evidentiary issues. For example, if an employer uses AI to track employee communications and discovers that an employee leaked documents to a journalist, that AI‑generated evidence may be contested as violating privacy rights under the GDPR or similar laws. The Court of Justice of the European Union in Meta Platforms Inc. v. Bundeskartellamt (Case C‑252/21, 2023) ruled that even automated processing must have a lawful basis under Article 6 GDPR, which affects the admissibility of such evidence in civil proceedings.

View Meta Platforms v. Bundeskartellamt (CJEU)

Chapter 5: Disputes Strategy and the Rule of Law – Adapting to AI-Driven Evidence

⬅ Back to Table of Contents

Developing an AI Evidence Strategy from Intake to Trial

An effective disputes strategy must incorporate AI considerations at every stage. During intake, counsel should assess whether adverse evidence may be AI‑generated or whether key documents were processed by an opponent’s AI system. During discovery, parties should request production of: (1) logs of AI‑assisted review (to challenge completeness), (2) model cards and training data for any AI system that generated or analysed evidence, and (3) chain‑of‑custody records for algorithmically produced outputs.

At the expert stage, parties may need duelling forensic experts: one to validate the authenticity of proffered AI evidence, another to attack it. Tribunal appointments of neutral AI experts are becoming more common in complex tech disputes.

At trial or hearing, counsel can use AI‑powered real‑time transcription and courtroom analytics, but must also anticipate objections that such tools themselves produce inadmissible work product or invade privilege.

Rule of Law Principles for AI Evidence

The rule of law demands that evidence be reliable, procedurally fair, and subject to scrutiny. Applying these principles to AI evidence requires:

• Transparency: The party offering AI‑generated evidence must disclose the model, version, inputs, and any significant limitations.
• Adversarial Testing: The opposing party must have an opportunity to inspect the AI system or its outputs under protective conditions if confidentiality is claimed.
• Human Oversight: A human witness must ordinarily sponsor AI evidence, explaining how it was produced and why it is reliable.
• Proportionality: The cost of validating or challenging AI evidence should not exceed the amount in controversy, but in high‑value disputes, thorough forensic examination is required.
• Remedies for Abuse: If a party knowingly introduces fabricated AI evidence or fails to preserve audit trails, sanctions should be available, including adverse inferences and cost awards.

These principles echo the Sedona Conference Commentary on Artificial Intelligence and Evidence (2023), which provides practical guidance for courts and arbitrators on managing AI‑related evidentiary issues.

Looking Ahead – What Legal Professionals Must Learn Now

The pace of AI development means that legal professionals cannot wait for settled case law. They must proactively understand basic AI concepts: how large language models generate text, what metadata is preserved in common file formats, how predictive coding algorithms classify documents, and where vulnerabilities exist in watermarking and detection.

Continuing legal education (CLE) programmes should include hands‑on workshops: one on using AI for e‑discovery, another on cross‑examining AI expert witnesses, and a third on drafting discovery requests for AI systems. Law firms and dispute resolution institutions should establish AI evidence task forces to develop standard protocols and model orders.

London International Disputes Week 2026 and similar forums provide opportunities to share emerging practices. The future of evidence is already here—it is multimodal, machine‑generated, and always accompanied by a supply chain. Mastering it is essential for the rule of law.

Related Topics

⬅ Back to Table of Contents

The following topics expand this playbook into broader areas of AI, law, and dispute resolution.

• Algorithmic Accountability in Judicial Decisions – How courts are reviewing AI‑assisted sentencing and risk assessments.
• E‑Discovery and Predictive Coding Standards – Best practices for technology‑assisted review under Fed. R. Civ. P. 26 and 34.
• Deepfake Litigation and Anti‑Deepfake Legislation – Emerging laws and forensic countermeasures.
• AI as a Witness – Legal personhood, chatbot testimony, and hearsay exceptions.
• Data Privacy and Cross‑Border Evidence – GDPR, Schrems II, and restrictions on transferring evidence containing personal data.
• Human Rights Due Diligence in AI Procurement – Supply chain audits under UN Guiding Principles and OECD Guidelines.
• Mass Arbitration and AI Case Management – Innovations from AAA, CPR, and Hague Rules.
• Generative AI Legal Ethics – Risks of using ChatGPT to draft pleadings or generate evidence.

FAQ

⬅ Back to Table of Contents

Can AI‑generated content be admitted as evidence in court or arbitration?

Yes, provided it meets standard admissibility requirements: relevance, authenticity, and not otherwise excluded (e.g., hearsay, privilege). Many jurisdictions treat AI‑generated outputs as “machine statements” that may be admitted if a human witness explains the process and reliability. However, deepfakes that cannot be authenticated are likely to be excluded.

What is predictive coding (TAR) and how does it relate to discovery?

Predictive coding (Technology‑Assisted Review) uses machine learning to rank documents by relevance to legal issues. A human attorney reviews a small sample, the AI learns from that coding, then applies it to millions of documents. It is widely accepted in US federal courts (e.g., Da Silva Moore v. Publicis Groupe, 2012) and increasingly in international arbitration.

How can I challenge the authenticity of an AI‑generated exhibit?

Challenge the chain of custody, ask for the model and training data, hire a forensic expert to look for statistical anomalies (e.g., inconsistent audio frequency patterns, pixel‑level artifacts in images), and demand the opposing party preserve all audit logs. If they fail to preserve, request an adverse inference.

Does using AI for e‑discovery waive privilege?

Not automatically. Under US law (Fed. R. Evid. 502), the use of AI for privilege review does not waive privilege as long as reasonable steps are taken to prevent disclosure. However, providing an opponent access to the AI system itself or to the coding decisions could waive work product protection. Protective orders are essential.

What human rights issues arise in the AI supply chain?

Key issues include: mining of rare earth minerals in conflict zones, labour conditions for data annotators (often paid piecework), use of sweatshop labour for content moderation, mass surveillance via training data scraping, and lack of consent for biometric data. These can give rise to litigation under consumer protection, human rights, or supply chain due diligence laws.

References

⬅ Back to Table of Contents

Verified source links (embedded):

• London International Disputes Week 2026 – Official Site (LIDW)
• American Society of International Law (ASIL) 2026 Annual Meeting – “The Great AI Race”
• State v. Loomis, 881 N.W.2d 749 (Wis. 2016) – Justia
• United States v. Gitarts, No. 20-cr-145 (S.D.N.Y. 2021) – Casetext
• Draft EU AI Liability Directive (2022/0303(COD)) – European Parliament
• In re Clearview AI, Inc., Consumer Privacy Litigation (N.D. Ill. 2024) – ClassAction.org
• Meta Platforms Inc. v. Bundeskartellamt, Case C-252/21 (CJEU 2023)
• UN Guiding Principles on Business and Human Rights (UNGPs)
• Sedona Conference Commentary on Artificial Intelligence and Evidence (2023)
• Equifax Data Breach Settlement – US DOJ (mass claims context)
• IBA Rules on the Taking of Evidence in International Arbitration (2020)