Civil Law And Machine Learning Evidence Evaluation And Automated Fact Finding In Uae Courts .

1. Introduction

Machine learning evidence evaluation and automated fact-finding represent an emerging development in the UAE civil justice system. These technologies involve using artificial intelligence (AI), statistical models, predictive algorithms, natural-language processing and automated data-analysis systems to assist courts in determining disputed facts and assessing evidence.

In traditional civil litigation, judges evaluate documentary evidence, witness testimony, expert opinions, electronic records, contractual documents and other material before reaching factual conclusions. Machine learning can assist this process by identifying patterns, detecting inconsistencies, analysing large volumes of documents and estimating the probability of competing factual explanations.

However, machine-generated conclusions are not automatically equivalent to judicial findings of fact. Under UAE law, the court must retain responsibility for evaluating evidence, ensuring procedural fairness and explaining its decision.

The central legal question is therefore:

To what extent can UAE courts use machine learning to evaluate evidence and establish facts without compromising judicial independence, due process and the right to challenge evidence?

The subject must be approached carefully: UAE legislation recognises electronic evidence and regulates expert evidence, but there is not yet an established body of reported UAE judgments specifically authorising machine learning systems to determine disputed civil facts autonomously.

2. Legal framework governing evidence in UAE civil proceedings

Several legislative instruments are relevant.

A. Federal Decree-Law No. 35 of 2022 on Evidence in Civil and Commercial Transactions

This is a principal legislative foundation for civil and commercial evidence in the UAE.

Its framework addresses:

Written and documentary evidence.

Electronic evidence and electronic records.

Admission and denial of evidence.

Evidentiary value of documents.

Witness testimony.

Expert evidence.

Judicial inspection and other evidentiary procedures.

Machine learning systems may assist in processing electronic evidence, but the admissibility and evidential weight of the underlying material must be assessed under the applicable statutory rules.

For example, an algorithm analysing thousands of commercial emails may identify communications relevant to a contractual dispute. The court must nevertheless examine the authenticity, integrity, origin and relevance of those communications.

B. Federal Decree-Law No. 42 of 2022 on Civil Procedure

The Civil Procedure Law provides the procedural framework for UAE civil litigation.

Its relevance includes:

Judicial management of evidentiary proceedings.

Appointment and supervision of experts.

Submission and examination of evidence.

Procedural opportunities for parties to respond.

Judicial assessment of disputed facts.

Reasoned judicial decisions.

An algorithm can support judicial case management and evidence analysis, but it cannot independently exercise judicial authority merely because its output is technically sophisticated.

C. Federal Decree-Law No. 46 of 2021 on Electronic Transactions and Trust Services

This legislation supports the legal recognition of electronic transactions, electronic signatures and trust services.

It is important when machine learning systems analyse:

Digitally signed contracts.

Electronic correspondence.

Electronic authentication records.

Business transaction histories.

Digital audit trails.

The electronic origin and integrity of evidence remain important considerations.

D. Federal Decree-Law No. 45 of 2021 on the Protection of Personal Data

Machine learning evidence evaluation may involve extensive processing of personal information.

Relevant concerns include:

Lawful processing.

Data security.

Purpose limitation.

Personal-data protection.

Appropriate safeguards.

The application of this legislation must also account for its statutory scope and exclusions, including particular government and other excluded processing activities.

E. UAE artificial intelligence governance

UAE AI governance policies and institutional initiatives encourage responsible AI adoption. Nevertheless, policy support for AI does not itself establish that an algorithm's factual conclusions are admissible or binding in civil proceedings.

3. Meaning and operation of machine learning evidence evaluation

Machine learning evidence evaluation refers to the application of computational models to analyse evidence and assist in identifying relevant factual conclusions.

Unlike conventional software, which follows predetermined instructions, machine learning systems identify patterns from training data and generate predictions or classifications.

A. Automated document analysis

A machine learning system can examine:

Commercial contracts.

Emails and electronic correspondence.

Bank statements.

Financial transaction records.

Corporate records.

Digital signatures.

Invoices and payment histories.

For example, in a commercial dispute involving alleged breach of contract, an algorithm may identify emails indicating that one party received notice of a contractual obligation.

The court can use this analysis to locate relevant documents, but the documents themselves must be examined in their evidentiary context.

B. Pattern recognition

Pattern recognition involves identifying relationships between different pieces of evidence.

Examples include:

Identifying repeated financial transactions.

Detecting inconsistencies in commercial records.

Comparing signatures.

Identifying unusual transaction sequences.

Detecting duplicated or manipulated documents.

Establishing chronological relationships between communications.

C. Predictive evidence evaluation

Predictive models may estimate the likelihood of particular factual explanations.

For example, a system analysing construction records might estimate whether project delays were associated with late material delivery, contractor performance or design changes.

However, statistical likelihood must not be confused with the legal standard of proof.

In civil proceedings, a party generally must establish its case on the applicable civil standard of proof. A machine-generated probability cannot independently determine whether that standard has been satisfied.

D. Automated fact classification

Machine learning systems may classify evidence into categories:

CategoryAutomated function
AuthenticityIdentify potential document manipulation
RelevanceLocate material connected with disputed issues
ChronologyReconstruct events in sequence
ConsistencyIdentify contradictory statements
Financial evidenceDetect transaction patterns
Contractual evidenceIdentify obligations and communications
Expert evidenceOrganise technical findings

Such classification is useful for managing large evidentiary records.

Nevertheless, classification is not equivalent to a final judicial finding.

4. Legal admissibility of machine-generated evidence

An important distinction must be made between electronic evidence and machine-generated conclusions.

Article 53 of Federal Decree-Law No. 35 of 2022 defines electronic evidence broadly through electronically generated, stored, extracted, copied, transmitted or received information.

Article 54 includes electronic records, electronic correspondence, electronic signatures, electronic media and other electronic evidence.

Article 55 subjects electronic evidence to the relevant rules governing documentary evidence.

Articles 56–59 address the evidentiary value of different forms of electronic evidence.

Article 60 deals with production of electronic evidence in its original format or through other electronic means.

Article 62 allows the court to assess the evidentiary value of electronic material when its validity cannot be verified for reasons not attributable to the litigants.

UAE Legislation+2

 

Four legal tests

1. Authenticity test

The court must determine whether the underlying electronic material genuinely originates from the claimed source.

2. Integrity test

The court must examine whether the material has been altered, manipulated or corrupted.

3. Reliability test

The court must consider whether the machine learning process is technically reliable and suitable for the particular factual issue.

4. Procedural fairness test

The opposing party must have a meaningful opportunity to challenge the underlying evidence, methodology and conclusions.

A machine-generated report that cannot be independently examined should not automatically be treated as conclusive evidence.

5. Judicial use of machine learning in UAE civil litigation

Machine learning can assist UAE courts in several important stages.

A. Pre-trial evidence management

Before trial, AI systems can organise large collections of documents.

They can:

Identify relevant evidence.

Group documents by subject.

Detect duplicate records.

Identify missing documentation.

Generate chronological summaries.

Identify potential contradictions.

This can reduce the time required to examine complicated commercial disputes.

B. Automated factual reconstruction

In contractual litigation, machine learning can reconstruct the sequence of events.

For example, a construction dispute may involve:

Original project commencement.

Design approval.

Material procurement.

Contractor correspondence.

Variations.

Delay notices.

Completion certificates.

An automated system can organise these events chronologically and identify potential causes of delay.

However, the court must distinguish correlation from causation.

C. Financial evidence evaluation

Machine learning can be particularly useful in banking and commercial disputes.

It can examine:

Banking transactions.

Payment instructions.

Loan repayment histories.

Corporate financial records.

Suspicious transfers.

Accounting inconsistencies.

For instance, an algorithm may identify transactions suggesting that funds were transferred through several related companies.

The resulting analysis may support an investigation into asset movement, but it does not independently prove fraud, beneficial ownership or unlawful intention.

D. Witness credibility assessment

This is one of the most legally sensitive applications.

AI systems may compare witness statements, identify inconsistencies and analyse differences between earlier and later accounts.

However, automated credibility scoring raises serious problems.

A witness's credibility cannot reliably be established merely through:

Facial expressions.

Voice patterns.

Speaking speed.

Emotional presentation.

Automated behavioural classification.

The court must consider the testimony in its full evidentiary context. Machine-generated credibility scores should not replace judicial assessment.

E. Automated expert assistance

Machine learning can support court-appointed experts in:

Digital forensic examination.

Handwriting and signature comparison.

Engineering calculations.

Accounting analysis.

Medical record review.

Data reconstruction.

The expert must explain the methodology and limitations of the system rather than simply presenting its output.

6. Role of court-appointed experts

Federal Decree-Law No. 35 of 2022 contains a specific framework for expert evidence.

Articles 109 onwards regulate the appointment and work of experts.

Article 109 allows the court or supervising judge to appoint one or more experts to provide opinions on technical matters requiring expertise.

Article 116 addresses the contents of expert reports.

Article 119 regulates the submission of expert reports and provides an opportunity for litigants to comment on the initial report.

Article 122 is particularly important:

An expert's opinion is not ordinarily binding on the court.

The court may depart from the expert's opinion.

If it does so, it must explain its reasons.

The parties may agree to accept an expert report, subject to the statutory qualification concerning public order.

Article 125 permits the court to rely on an expert report from another case while preserving the litigants' right to discuss its contents.

littdb.sfo2.cdn.digitaloceanspaces.com+1

 

Application to machine learning experts

A machine learning specialist appointed as an expert should explain:

The source of the training data.

The source of the evidence analysed.

The algorithm used.

The model's intended purpose.

The accuracy and error rates.

The possibility of bias.

The limitations of the results.

Whether the analysis can be independently reproduced.

An expert cannot avoid responsibility by claiming that the algorithm independently generated the conclusion.

7. Six important UAE-related judicial decisions

The following decisions are relevant by analogy to machine learning evidence, digital authentication, expert methodology, evidentiary inference and AI reliability. They do not collectively establish that UAE courts have approved autonomous AI fact-finding. DIFC and ADGM decisions must also be distinguished from judgments of the UAE onshore federal and local courts.

Case 1. Arabyads Holding Limited v Gulrez Alam Marghoob Alam [2025] ADGMCFI 0032

Direct AI-related authority

This ADGM Court of First Instance decision is particularly important for the treatment of AI-generated material in litigation.

The dispute involved legal submissions containing authorities that did not exist or were inaccurately cited and which bore the hallmarks of AI-generated hallucinations.

The court imposed wasted costs of AED 282,508 against the relevant law firm.

Legal significance:

AI-generated material requires verification.

Technological assistance does not eliminate professional responsibility.

Unverified machine-generated information should not be treated as reliable.

Human accountability remains essential in judicial proceedings.

Application: A party submitting an AI-generated evidentiary report should be required to establish the authenticity and reliability of its contents.

Case 2. Bond Interior Design LLC v Tr88house Restaurant and Entertainment Center LLC [2023] DIFC TCD 001

Black-box software analogy

This construction dispute involved technical expert evidence and the use of Primavera project-management software.

The court examined software-generated calculations and the limitations of expert evidence where the expert could not adequately explain or independently verify the conclusions produced by the software.

The case is particularly relevant to machine learning because sophisticated software can produce outputs without necessarily providing an adequately demonstrated explanation of how those outputs were reached.

Legal significance:

Software-generated calculations require appropriate verification.

Experts must understand the methodology supporting their conclusions.

Automated outputs should not be accepted merely because they appear technically sophisticated.

The court retains responsibility for evaluating the evidence.

Application: An AI system identifying the cause of construction delays should disclose the data and analytical methodology supporting its conclusion.

Case 3. ICICI Bank Limited v Bavaguthu Raghuram Shetty [2022] DIFC CFI 034

Digital authentication authority

This banking dispute concerned personal guarantees and facility agreements involving handwritten and electronically reproduced signatures.

The court considered handwriting expert evidence and whether electronic or copied signatures had been applied with the defendant's authority.

The distinction between the authenticity of an underlying signature and the authority for its electronic application was important.

The judgment demonstrates that technical examination of a digital signature cannot necessarily establish all the surrounding legal facts.

Legal significance:

Technical authenticity and legal authorisation are different questions.

Expert opinions must be assessed within their proper technical limits.

Documentary evidence can corroborate other evidence.

The court must consider the complete evidentiary record.

Application: A machine learning system may identify a signature as matching a known signature, but it cannot automatically establish the signatory's consent or authority.

Case 4. SBM Bank (Mauritius) Ltd v Renish Petrochem FZE and Hiteshkumar Chinubhai Mehta [2022] DIFC CA 011

This case concerned banking and financial disputes involving complex documentary and evidentiary issues.

The Court of Appeal considered challenges to the first-instance decision and the treatment of evidence in the context of the underlying financial transactions.

Legal significance:

The decision is useful when considering the relationship between factual findings, documentary material and appellate review.

Application to machine learning:

A predictive system may identify suspicious financial activity, but the court must examine the underlying transaction documents, surrounding circumstances and competing explanations before determining liability.

The model's classification cannot replace the judicial assessment of the complete evidence.

Case 5. Fidel v Felecia and Faraz [2015] DIFC CA 002

This decision concerned the admission of expert evidence relating to UAE public policy and the treatment of non-DIFC UAE law before the DIFC Courts.

The Court of Appeal examined the discretion available under Article 50 of the DIFC Courts Law concerning applicable evidentiary rules.

It rejected the argument that expert evidence must invariably be required to establish non-DIFC UAE law.

Legal significance:

Expert evidence is not automatically necessary for every disputed issue.

The court must determine what kind of expertise is actually required.

Judicial competence and procedural context matter.

Evidence rules should be applied appropriately to the circumstances.

Application: A machine learning expert may assist with technical questions, but the interpretation of legal rules remains a judicial responsibility.

Case 6. Techteryx Ltd v Aria Commodities DMCC and Others [2025] DIFC DEC 001

Digital financial evidence

This case involved a substantial dispute concerning reserves backing the TrueUSD stablecoin and claims relating to the transfer and ownership of approximately USD 456 million.

The proceedings involved proprietary and worldwide freezing injunctions and questions concerning the tracing and preservation of financial assets.

Legal significance:

The dispute illustrates the complexity of modern digital financial transactions and the importance of analysing transaction records, asset movements and documentary evidence.

Application to machine learning:

Machine learning may assist in:

Identifying transaction chains.

Analysing financial movements.

Detecting irregular transfers.

Identifying relationships between accounts.

Organising evidence relating to digital assets.

However, a detected transaction pattern does not independently establish ownership, fraud or beneficial entitlement.

Oxford Institute of Technology and Justice+5

8. Fundamental legal challenges

A. Algorithmic bias

Machine learning systems may reproduce biases contained in their training data.

For example, a system trained on historical commercial disputes may disproportionately associate particular business practices or transaction patterns with fraudulent conduct.

This can produce inaccurate conclusions.

Legal concern: Evidence must be evaluated objectively, without unjustified discrimination or prejudice.

B. Lack of algorithmic transparency

Some machine learning systems operate as black boxes.

Their internal processes may be difficult to understand, even for technically qualified users.

This creates difficulties when a litigant wishes to challenge:

The training dataset.

The model architecture.

The classification process.

The probability assigned to a conclusion.

The reasons for rejecting alternative explanations.

A court should be particularly cautious when a party cannot meaningfully challenge the methodology behind a decisive algorithmic conclusion.

C. Data manipulation

Machine learning results depend on the quality of the input data.

If the underlying records are manipulated, incomplete or incorrectly attributed, the algorithm may generate misleading results.

For this reason, digital chain of custody, metadata preservation and forensic verification are important.

D. Violation of procedural fairness

A party may be disadvantaged if the opposing party introduces an algorithmic report without disclosing its methodology.

Fair proceedings require a meaningful opportunity to:

Examine the underlying material.

Understand the analytical process.

Challenge the expert's assumptions.

Present contrary expert evidence.

Question the reliability of the conclusion.

E. Judicial independence

A major constitutional and institutional concern arises when judges become overly dependent on automated recommendations.

The final decision must remain the responsibility of the legally competent judicial authority.

An algorithm should not become an undisclosed substitute for judicial reasoning.

9. Right to challenge machine-generated evidence

A party opposing machine learning evidence should be able to raise objections concerning:

Ground of objectionLegal concern
Data authenticityWhether the original records are genuine
Data integrityWhether information was altered
Training biasWhether the model produces discriminatory results
Model reliabilityWhether the methodology is scientifically dependable
ExplainabilityWhether the conclusion can be understood
ReproducibilityWhether another expert can test the result
RelevanceWhether the analysis addresses an actual disputed issue
Procedural fairnessWhether the opposing party can challenge it

A court may require further expert examination where the reliability of an algorithmic conclusion is genuinely disputed.

10. DIFC and ADGM approaches

The UAE has several distinct judicial systems, and the distinction is important.

A. UAE onshore courts

Federal Decree-Law No. 35 of 2022 provides the principal federal framework for civil and commercial evidence.

The law recognises electronic evidence and regulates court-appointed experts.

It does not establish a general statutory system under which machine learning outputs automatically determine disputed facts.

B. DIFC Courts

The DIFC Courts operate under their own legal and procedural framework.

Their decisions concerning electronic signatures, expert evidence and technical software are useful for understanding the evidentiary challenges of algorithmic analysis.

The DIFC's AI-related practitioner guidance also emphasises transparency and verification of AI-generated material.

C. ADGM Courts

The ADGM Courts operate under their own common-law-based legal framework.

The decision in Arabyads Holding Limited demonstrates the importance of verifying AI-generated material before using it in court proceedings.

These judicial systems can provide useful comparative guidance, but their decisions should not automatically be treated as binding interpretations of federal UAE evidence legislation.

11. Proposed framework for responsible automated fact-finding in UAE courts

A sound legal framework could be developed around the following principles.

Stage 1 — Evidence collection

Collect original documents, electronic records, transaction histories and other relevant material.

Stage 2 — Digital authentication

Verify the source, integrity, metadata and chain of custody of the material.

Stage 3 — Machine learning analysis

Use algorithms to identify patterns, inconsistencies, relationships and potential factual explanations.

Stage 4 — Expert verification

Require qualified experts to explain the model, assumptions, accuracy and limitations.

Stage 5 — Adversarial examination

Allow the opposing party to examine and challenge the algorithmic evidence and expert conclusions.

Stage 6 — Independent judicial evaluation

The judge evaluates all evidence independently and determines the facts under the applicable legal standard.

Stage 7 — Reasoned judgment

The court explains the evidentiary basis of its findings and the weight, if any, given to machine-generated analysis.

12. Practical examples in UAE civil litigation

Type of disputePotential machine learning application
Construction litigationIdentifying causes and periods of project delay
Banking disputesDetecting irregular transactions
Contract litigationReconstructing contractual communications
Insurance disputesIdentifying inconsistencies in claims
Property disputesComparing registration and ownership records
Corporate litigationIdentifying related-party transactions
Employment disputesAnalysing employment communications
Digital asset disputesTracing blockchain transactions
Medical negligenceOrganising clinical records and identifying potential inconsistencies

These applications are most defensible when the technology assists evidence management and technical analysis rather than making the final determination of liability.

13. Recommendations for UAE legal reform

The UAE could strengthen the framework through several reforms.

1. Algorithmic evidence admissibility standards

Specific procedural standards should address when machine-generated reports may be admitted and how their evidential weight should be assessed.

2. Mandatory disclosure

Parties relying on material machine learning analysis should disclose the relevant methodology, source data and material limitations, subject to legitimate confidentiality and privacy protections.

3. Independent technical examination

Courts should have access to qualified independent experts capable of examining complex machine learning systems.

4. Human judicial control

No machine learning system should independently determine civil liability or replace the judge's responsibility to evaluate disputed facts.

5. Protection against algorithmic bias

Evidence evaluation systems should undergo appropriate testing for discriminatory outcomes and technical errors.

6. Preservation of digital evidence

Courts should establish reliable procedures for preserving original electronic material, metadata and audit trails.

7. Explainable judicial decisions

Where algorithmic analysis materially influences a factual finding, the judgment should explain the relevant evidentiary reasoning.

14. Conclusion

Machine learning evidence evaluation and automated fact-finding have significant potential to modernise civil litigation in the UAE.

They can assist courts in examining extensive documentary records, detecting financial irregularities, reconstructing contractual events, identifying inconsistencies and evaluating complicated technical evidence.

Federal Decree-Law No. 35 of 2022 already provides an important foundation through its recognition of electronic evidence and regulation of expert evidence.

The decisions in Arabyads, Bond Interior Design, ICICI Bank, SBM Bank, Fidel and Techteryx provide useful judicial guidance on AI reliability, software-generated conclusions, electronic authentication, expert evidence and digital financial records.

However, these decisions do not establish a general legal doctrine permitting autonomous machine adjudication.

The most appropriate approach is human-centred algorithmic adjudication: machines assist in identifying and organising facts, experts verify technical conclusions, litigants retain the ability to challenge evidence, and judges independently determine the legal consequences.

Ultimately, the legitimacy of automated fact-finding in UAE civil courts must depend on accuracy, transparency, procedural fairness, accountability and judicial independence.

15. Important legislation and case-law revision table

AuthorityPrincipal relevance
Federal Decree-Law 35/2022Civil and commercial evidence
Federal Decree-Law 42/2022Civil procedure
Federal Decree-Law 46/2021Electronic transactions
Federal Decree-Law 45/2021Personal data protection
Arabyads [2025] ADGMCFI 0032AI-generated material
Bond Interior Design [2023] DIFC TCD 001Software and expert reliability
ICICI Bank [2022] DIFC CFI 034Electronic signatures and expert evidence
SBM Bank [2022] DIFC CA 011Financial evidence and appellate review
Fidel [2015] DIFC CA 002Expert evidence and judicial discretion
Techteryx [2025] DIFC DEC 001Digital financial evidence

Central legal proposition: Machine learning may assist UAE courts in discovering, organising and analysing facts, but the authority to determine disputed facts and apply the law must remain with the court.

 

 

LawGratis provides pro bono legal aid and resources to empower communities. Join us in making justice reachable for everyone! Follow our community on Instagram, YouTube, Facebook, and LinkedIn.

LEAVE A COMMENT