Civil Law and UAE AI-Generated Evidence Admissibility Standards

1. Introduction

AI-generated evidence refers to evidence that is created, reconstructed, transformed, analysed, or materially processed using an artificial-intelligence system.

Examples include:

• AI-generated documents;
• AI-generated images;
• synthetic audio;
• deepfake videos;
• AI-generated emails;
• automated transcripts;
• AI translations;
• AI-created summaries;
• AI-enhanced photographs;
• AI-generated contracts;
• algorithmically reconstructed records;
• AI-assisted forensic analysis;
• outputs from large language models;
• machine-generated reports.

The central legal question is not simply:

“Was this evidence generated by AI?”

The correct question is:

“Is the evidence authentic, relevant, reliable, attributable, sufficiently verifiable, and legally capable of proving the fact for which it is tendered?”

This distinction is particularly important in the UAE because the Federal Decree-Law No. 35 of 2022 on Evidence in Civil and Commercial Transactions expressly recognizes electronic evidence. Article 53 broadly defines electronic evidence as evidence derived from data or information generated, stored, extracted, copied, transmitted, reported or received through information technology and retrievable in an understandable manner. Articles 54–62 then establish categories and evidentiary rules for electronic evidence.

The current UAE civil-law framework must also be read alongside the Federal Decree-Law No. 25 of 2025 promulgating the Civil Transactions Law, effective 1 June 2026.

2. AI-Generated Evidence vs Electronic Evidence

An important distinction must be made.

Electronic evidence

Electronic evidence is a broad statutory category.

It can include:

• emails;
• electronic records;
• electronic signatures;
• electronic correspondence;
• electronic communications;
• electronic media;
• other electronically generated information.

The Evidence Law expressly includes these categories.

AI-generated evidence

AI-generated evidence is a technologically newer subset or form of electronically generated material.

For example:

An ordinary email = electronic evidence.

But:

An AI-generated email or AI-generated reconstruction of an alleged communication = potentially AI-generated electronic evidence.

The fact that AI was involved does not automatically make the material inadmissible.

But it creates additional questions about authenticity and reliability.

3. Statutory Recognition of Electronic Evidence

The UAE Evidence Law provides an important starting point.

Article 53

Electronic evidence covers evidence derived from electronically generated, stored, extracted, copied, transmitted, reported or received data that can be retrieved in understandable form.

Article 54

Electronic evidence expressly includes:

1. electronic records;
2. electronic instruments;
3. electronic signatures;
4. electronic seals;
5. electronic correspondence, including emails;
6. modern means of communication;
7. electronic media; and
8. other electronic evidence. 

This final category is particularly important for emerging technologies.

It creates room for courts to evaluate new forms of electronic evidence without requiring the legislature to list every future technology individually.

4. Article 55: Same Evidentiary Framework

Article 55 provides that electronic evidence is subject to the provisions governing documentary evidence under the Evidence Law.

Therefore, electronic form alone does not determine evidentiary weight.

The court must still consider:

• authenticity;
• reliability;
• integrity;
• legal requirements;
• attribution;
• probative value.

This principle is highly relevant to AI-generated evidence.

5. Formal Electronic Evidence

Article 56 gives qualifying formal electronic evidence the same probative value as formal instruments when the statutory requirements are satisfied. It also addresses documents automatically generated by electronic systems of public entities or entities entrusted with public services.

This raises an important distinction:

Automatically generated electronic evidence

For example:

• government system records;
• official electronic certificates;
• authenticated system-generated records.

Generative-AI output

For example:

• ChatGPT-style generated text;
• AI-created image;
• synthetic voice;
• AI-generated contract.

The second category should not automatically receive the evidentiary status of an official automatically generated record.

6. Informal Electronic Evidence

Article 57 recognizes certain forms of informal electronic evidence as legally valid between transaction parties, including where the evidence:

• is issued according to applicable legislation;
• is generated through an electronic means specified in the relevant contract;
• or is generated through an authenticated or publicly available electronic means. 

This becomes important for:

• electronic contracting;
• automated business systems;
• digital communications;
• AI-assisted transactions.

7. Burden of Challenging Electronic Evidence

Article 58 is particularly significant.

Where electronic evidence falls within Articles 56 and 57, the party alleging its invalidity bears the burden of proving that allegation.

This means that the legal position is not necessarily:

“Electronic evidence is suspicious until proven genuine.”

The statutory framework may allocate the burden differently depending upon the category and circumstances.

For AI-generated material, however, questions of authentication and attribution may become central before the court can confidently give the material significant evidentiary weight.

8. Article 59: Probative Value

Article 59 provides that, except where Article 56 applies, electronic evidence generally has the same probative value as informal instruments under the Evidence Law.

This is important because:

Admissibility does not necessarily equal conclusiveness.

An AI-generated record may be admitted or considered but ultimately given little weight.

9. Article 60: Original Format

Article 60 permits electronic evidence to be produced in its original format or through another electronic means, and allows the court to request its contents in written form where the nature of the evidence permits.

For AI evidence, preserving the original electronic environment may be extremely important.

For example, instead of producing only:

“AI generated this document.”

a party should ideally preserve:

• original file;
• metadata;
• timestamps;
• source data;
• system logs;
• model/version information;
• relevant prompts;
• output;
• subsequent modifications.

10. Verification of AI-Generated Evidence

The central question is:

Can the court verify how the evidence came into existence?

A useful verification framework is:

Source

Where did the information originate?

System

What AI or technological system created or processed it?

Input

What information was provided to the system?

Processing

What transformation occurred?

Output

What exactly did the AI generate?

Modification

Was the output subsequently edited?

Attribution

Who caused or authorized the creation?

Integrity

Has the evidence remained unchanged?

Corroboration

Is it supported by independent evidence?

11. Authentication

Authentication means establishing that the evidence is what the party claims it to be.

For AI-generated evidence, authentication can be difficult.

Suppose a claimant submits an audio recording allegedly showing that a defendant made a contractual admission.

The defendant argues:

“That is an AI-generated voice.”

The court may need to consider:

• original recording;
• metadata;
• device information;
• recording history;
• forensic examination;
• voice analysis;
• witnesses;
• surrounding communications;
• chain of custody.

An AI-generated recording should not be treated as genuine merely because it sounds authentic.

12. Deepfakes

Deepfakes represent one of the most serious AI-evidence problems.

They can create:

• synthetic videos;
• synthetic voices;
• fabricated photographs;
• manipulated facial expressions;
• simulated statements.

A deepfake may appear visually and acoustically convincing.

Therefore, visual plausibility is not equivalent to authenticity.

Courts may need technical expert assistance.

13. AI-Generated Documents

AI can generate:

• contracts;
• invoices;
• letters;
• legal notices;
• corporate resolutions;
• emails;
• reports.

The court must distinguish between:

AI-generated document

Created by AI without necessarily representing an actual historical transaction.

AI-assisted document

Created by a person using AI but actually approved or signed by the relevant party.

Authentic electronically generated document

Created automatically by an established business or government system.

These categories can have very different evidentiary significance.

14. AI-Generated Email

Suppose a claimant produces an email stating:

“We accept your settlement proposal.”

The claimant says:

“The company's AI system sent this email.”

The court must determine:

1. Who authorized the AI system?
2. Was the system acting within its authority?
3. Was the email actually sent?
4. Was the account compromised?
5. Was the content subsequently altered?
6. Was the communication attributable to the company?
7. Did the recipient reasonably rely on it?

Thus, creation and attribution are separate questions.

15. AI Transcripts

AI transcription is increasingly common.

For example:

A hearing is recorded and AI converts the audio into text.

The transcript may be useful evidence, but the original audio can be more authoritative if a dispute arises.

Potential errors include:

• misheard words;
• incorrect names;
• omitted statements;
• punctuation errors;
• translation problems.

Therefore:

Audio recording → AI transcript → verification

is safer than treating the AI transcript as infallible.

16. AI Translation

AI translation can be extremely useful in UAE litigation because disputes may involve:

• Arabic;
• English;
• French;
• Chinese;
• Hindi;
• other languages.

However, legal terminology can be highly context-dependent.

A translation generated by AI may therefore require:

• qualified human verification;
• comparison with the original;
• certification where required.

The original document should not disappear merely because an AI translation is available.

17. AI-Enhanced Photographs

AI can enhance images by:

• increasing resolution;
• removing noise;
• reconstructing missing pixels;
• sharpening objects;
• restoring damaged photographs.

This creates an important distinction between:

Enhancement

Improving visibility of existing information.

Reconstruction

Generating information that was not actually captured.

The more an AI system reconstructs missing information, the more carefully the court should examine whether the resulting image represents historical evidence or algorithmic inference.

18. AI-Generated Legal Research

An AI system may produce a case-law summary.

For example:

“The UAE Court of Cassation decided that X is always unlawful.”

Before relying on it, the adjudicator or lawyer should verify:

• actual existence of the case;
• case number;
• court;
• date;
• precise holding;
• statutory provision;
• factual context.

AI hallucination is particularly dangerous in legal research.

19. AI-Generated Expert Analysis

AI may assist an expert in:

• financial calculations;
• engineering analysis;
• medical image processing;
• valuation;
• damages calculation.

But the court must distinguish between:

expert's independent professional opinion

and

AI output merely reproduced by the expert.

Where AI materially contributes to the expert's conclusion, the methodology may need to be disclosed sufficiently for meaningful challenge.

20. AI Evidence and Expert Evidence

Expert evidence can become crucial where authenticity is technically disputed.

An expert may investigate:

• metadata;
• digital signatures;
• file hashes;
• system logs;
• image manipulation;
• audio synthesis;
• model outputs;
• data provenance.

The expert's role is not to decide the legal issue.

The court remains responsible for determining the evidentiary consequences.

21. Chain of Custody

AI-generated or AI-processed evidence requires a reliable chain of custody.

A simplified chain is:

Creation → Collection → Preservation → Transfer → Analysis → Presentation

At every stage, the party should ideally be able to demonstrate:

• who possessed the evidence;
• when it was transferred;
• whether it was modified;
• what tools were used;
• whether the original remains available.

A broken chain of custody may reduce evidentiary weight.

22. Reliability

Reliability asks:

Can the court reasonably trust the process that produced the evidence?

Relevant factors include:

• system integrity;
• error rate;
• reliability of source data;
• known AI limitations;
• model version;
• reproducibility;
• expert validation;
• independent corroboration.

A highly sophisticated AI system can still produce an unreliable result.

23. Relevance

Evidence must relate to an issue in dispute.

For example:

An AI-generated personality assessment of a claimant may be technologically sophisticated but irrelevant to whether a contract was breached.

Therefore:

Technological sophistication ≠ legal relevance.

24. Probative Value

Even relevant AI evidence may have limited probative value.

Consider:

AI estimates a 92% probability that a particular event occurred.

That does not automatically mean the legal fact is established.

The court must consider:

• methodology;
• assumptions;
• alternative explanations;
• underlying data;
• legal standard of proof;
• corroborating evidence.

25. AI and Legal Causation

AI may provide probability-based analysis.

But civil law requires a legal determination of causation.

For example:

AI predicts that a defective construction design had an 85% probability of causing the structural failure.

The court still needs to determine whether the legal requirements for causation and civil responsibility are satisfied.

Therefore:

Algorithmic probability is evidence; it is not automatically legal causation.

26. AI Evidence and Privacy

AI evidence may contain personal data.

Examples include:

• biometric information;
• faces;
• voices;
• medical records;
• financial information;
• communications.

The court must therefore balance evidentiary needs with applicable privacy and data-protection requirements.

27. AI Evidence and Trade Secrets

AI providers may resist disclosure of:

• source code;
• model weights;
• proprietary algorithms;
• training methods.

The opposing party may argue:

“Without seeing the model, I cannot challenge the evidence.”

This creates a tension between:

evidentiary fairness

and

commercial confidentiality.

A practical approach may involve:

• independent expert examination;
• restricted disclosure;
• confidentiality arrangements;
• controlled inspection;
• disclosure of methodology rather than proprietary source code.

28. AI Evidence in Arbitration

The UAE's arbitration environment creates similar issues.

AI-generated evidence may be used in:

• construction arbitration;
• commercial arbitration;
• investment disputes;
• technology disputes.

The tribunal should consider:

• authenticity;
• reliability;
• equal opportunity to challenge;
• confidentiality;
• procedural fairness.

AI cannot deprive a party of its opportunity to challenge evidence.

29. Six Important UAE Case-Law Authorities

Important qualification: UAE reported jurisprudence specifically deciding the admissibility of generative-AI-created evidence, deepfakes or foundation-model outputs remains limited. The following authorities therefore provide foundational or analogical principles concerning electronic evidence, evidentiary evaluation, expert evidence, reasoning and proof. They should not be misrepresented as direct AI-evidence precedents.

Case 1 — UAE Federal Supreme Court / UAE Federal Judicial Electronic-Evidence Jurisprudence

UAE courts have recognized the evidentiary significance of electronic records and electronic means where the applicable statutory requirements concerning authenticity and reliability are satisfied.

Relevance

This is the fundamental legal bridge between traditional evidence and AI-generated electronic evidence.

AI-generated material should therefore be evaluated within the statutory electronic-evidence framework rather than automatically excluded merely because it was produced electronically.

Principle:
Electronic form does not by itself destroy evidentiary value; authenticity and statutory requirements remain important.

30. Case 2 — UAE Federal Supreme Court, Civil Appeal No. 79/2020

This case concerned admission and proof, including the legal significance of a party's admission and the court's treatment of evidentiary material.

Relevance to AI

An AI-generated statement cannot automatically be treated as a human admission.

The court must establish:

• who generated it;
• who authorized it;
• whether it is attributable to the alleged party;
• whether the underlying communication is authentic.

Principle:
Evidence derives legal significance from its proper attribution and evidentiary character, not merely from its existence in documentary form.

31. Case 3 — UAE Federal Supreme Court, Commercial Appeal No. 215/2020

This authority concerns the court's treatment of expert reports and evidence.

The court emphasized that a judgment cannot simply rely mechanically upon an expert report without adequately addressing the reasoning and relevant defenses.

Relevance to AI

An AI-generated forensic report should not automatically become the court's reasoning.

If an AI system produces:

“The document is authentic with 97% probability,”

the court must still evaluate the underlying methodology and competing evidence.

Principle:
Expert or technical material assists the court; it does not eliminate the court's obligation to independently assess the evidence.

32. Case 4 — UAE Federal Supreme Court, Penal Cassation No. 1422/2022

The court emphasized that evidence supporting a judicial conclusion must be sufficiently probative and that the judgment should demonstrate that the court examined and assessed the evidence.

Relevance

This is highly relevant to AI-generated evidence.

A court should not simply state:

“The AI system determined that the document was authentic.”

The judgment should demonstrate why the evidence was accepted and what evidentiary weight it received.

Principle:
AI cannot substitute for judicial evaluation and reasoning.

33. Case 5 — UAE Federal Supreme Court, Penal Cassation No. 660/2023

The court emphasized the trial court's ability to evaluate the evidence presented, provided its conclusions are logically supported by evidence in the record, and stressed the importance of addressing material defenses.

Relevance

If a party alleges:

“This video is an AI-generated deepfake,”

the court should meaningfully address the challenge rather than simply relying on the appearance of the video.

Principle:
A material challenge to evidence must receive proper judicial consideration.

34. Case 6 — UAE Federal Supreme Court, Penal Cassation No. 1093/2019

The court recognized the trial court's authority to assess and weigh evidence and adopt evidence it finds sufficiently reliable and probative.

Relevance

AI-generated evidence should therefore be treated as material whose probative value must be assessed, rather than as automatically conclusive evidence.

Principle:
The court retains responsibility for evaluating the reliability and weight of evidence.

35. Case 7 — UAE Federal Supreme Court, Penal Cassation No. 1523/2022

The court reiterated that the trial court has broad authority to assess evidence and establish the factual relationship between the accused and the alleged conduct, provided the inference is sound and logically supported.

Relevance

This principle is useful where AI is used to reconstruct or analyse an event.

The AI's conclusion remains one evidentiary input.

The court must determine whether the inference is actually supported by the record.

Principle:
Algorithmic inference cannot replace judicial assessment of factual and evidentiary sufficiency.

36. Case 8 — UAE Federal Supreme Court, Penal Cassation No. 445/2022

The court emphasized the trial court's authority to assess evidence and rely on evidence it considers sufficient where its conclusion is properly supported.

Relevance

The case reinforces the distinction between:

admission of evidence

and

weight assigned to evidence.

An AI-generated file may be considered but still receive little weight if its authenticity or reliability is doubtful.

37. Six Core Admissibility Requirements

A practical UAE AI-evidence framework can therefore be formulated around six questions.

1. Relevance

Does the evidence relate to a fact in issue?

2. Authenticity

Is it genuinely what the party claims it is?

3. Integrity

Has it remained unaltered or can alterations be reliably identified?

4. Attribution

Can the evidence be legally connected to the person or event alleged?

5. Reliability

Was the method used to create or process it sufficiently dependable?

6. Probative value

How much weight should the court give it?

38. AI-Generated Evidence Admissibility Matrix

39. AI-Assisted vs AI-Generated Evidence

This distinction is extremely important.

AI-assisted evidence

A human creates the evidence and AI helps process it.

Example:

A human takes a photograph and AI improves its resolution.

AI-generated evidence

AI creates substantive content that did not previously exist.

Example:

AI generates a photograph depicting an event that never occurred.

The second category requires significantly greater scrutiny.

40. AI Reconstruction

Suppose CCTV footage is damaged.

An AI system reconstructs missing frames.

The resulting material may be useful, but the court should distinguish:

Original captured information

from

algorithmically inferred information.

The reconstructed portion should not necessarily be presented as though it were an original recording.

41. Deepfake Detection

When authenticity is challenged, courts may consider:

• metadata;
• cryptographic signatures;
• original device;
• file history;
• forensic analysis;
• compression patterns;
• audio inconsistencies;
• image artifacts;
• independent recordings;
• witness testimony.

No single detection method should necessarily be treated as infallible.

42. Burden of Proof

AI evidence may create complicated burden-of-proof questions.

For example:

Party A submits an electronically authenticated document.

Party B alleges:

“AI generated this document.”

The court must determine whether the statutory presumption or evidentiary status applies and who bears the burden of challenging authenticity.

Article 58 of the Evidence Law specifically places the burden of proving invalidity upon the litigant alleging invalidity in the circumstances covered by Articles 56 and 57.

The precise burden will therefore depend on the category of evidence and applicable law.

43. AI Evidence and Judicial Discretion

The court should distinguish three stages:

Stage 1 — Can the material be considered?

This is the admissibility/legality question.

Stage 2 — Is it authentic and reliable?

This concerns evidentiary quality.

Stage 3 — How much weight should it receive?

This concerns probative value.

Therefore:

Admissible does not mean decisive.

44. AI Evidence and Civil Liability

AI-generated evidence can itself become the basis for a civil claim.

Example:

A company submits an AI-generated image falsely showing that a contractor caused property damage.

The contractor suffers reputational and financial harm.

Possible issues include:

• wrongful conduct;
• authenticity;
• bad faith;
• causation;
• damage;
• attribution;
• compensation.

The use of AI may therefore create liability in addition to evidentiary questions.

45. AI Evidence and Abuse of Rights

A party should not manufacture AI evidence merely to obtain an improper litigation advantage.

For example:

• creating fake communications;
• generating fabricated invoices;
• producing synthetic photographs;
• manipulating evidence;
• creating false witness recordings.

If discovered, such conduct could have serious procedural and substantive consequences.

The UAE doctrine of abuse of rights provides an additional conceptual framework where procedural or substantive rights are exercised improperly.

46. AI Evidence and Good Faith

Good faith is relevant to the submission and use of evidence.

A party should not deliberately present synthetic evidence as though it were an authentic historical record.

Similarly, lawyers and experts should exercise appropriate care when relying upon AI-generated material.

47. AI Evidence in Commercial Litigation

Commercial disputes may increasingly involve:

• AI-generated invoices;
• automated contracts;
• synthetic business communications;
• AI financial forecasts;
• algorithmic trading records.

Courts should focus on:

source + integrity + attribution + reliability + corroboration.

48. AI Evidence in Construction Disputes

Construction disputes are especially suitable for AI-assisted evidence analysis because they contain enormous quantities of data.

AI can examine:

• project schedules;
• emails;
• variation orders;
• invoices;
• photographs;
• site records.

But an AI-generated delay analysis is not automatically legally conclusive.

The court must still determine:

• contractual responsibility;
• causation;
• concurrent delay;
• mitigation;
• damages.

49. AI Evidence in Employment Disputes

AI-generated evidence may include:

• recruitment scores;
• employee performance analytics;
• automated communications;
• workplace monitoring records.

The court may need to establish:

• how the data was collected;
• whether it was altered;
• whether the system accurately represented the employee's conduct;
• whether human review occurred.

50. AI Evidence in Family and Personal Matters

AI-generated evidence presents particularly serious concerns in disputes involving:

• family communications;
• photographs;
• audio recordings;
• social-media messages;
• identity;
• reputation.

Synthetic evidence can cause significant personal harm.

Courts should therefore apply particularly careful scrutiny to provenance and authenticity.

51. Recommended UAE Court Procedure

A useful procedural approach would be:

Step 1 — Identify the evidence

Is it electronic, AI-assisted or AI-generated?

Step 2 — Identify its source

Who or what created it?

Step 3 — Preserve the original

Maintain original files and metadata.

Step 4 — Establish authenticity

Verify the provenance.

Step 5 — Examine integrity

Determine whether it was altered.

Step 6 — Obtain technical expertise where necessary

Use qualified experts.

Step 7 — Permit adversarial challenge

Allow the opposing party to contest authenticity and methodology.

Step 8 — Determine evidentiary weight

The court independently assesses probative value.

52. Proposed AI Evidence Audit Trail

A robust AI-generated evidence package should ideally contain:

Original source
↓
Date/time
↓
Device/system
↓
AI tool/model
↓
Input data
↓
Processing instructions
↓
Output
↓
Modifications
↓
Human verification
↓
Final evidence

This substantially improves transparency and reliability.

53. Major Challenges

The UAE legal system will increasingly need to address:

1. deepfakes;
2. synthetic witnesses;
3. AI-generated documents;
4. AI-generated signatures;
5. synthetic voice recordings;
6. AI reconstruction;
7. AI hallucinations;
8. manipulated metadata;
9. model opacity;
10. trade-secret objections;
11. cross-border AI systems;
12. evidentiary burden allocation;
13. expert methodology;
14. chain of custody;
15. AI-generated legal authorities.

54. Key Legal Principle

The most important principle is:

AI generation does not automatically make evidence inadmissible, and electronic form does not automatically make AI-generated evidence reliable.

The court should separately examine:

Relevance → Authenticity → Integrity → Attribution → Reliability → Probative Value

This is consistent with the UAE Evidence Law's treatment of electronic evidence and the broader UAE judicial approach requiring courts to evaluate evidence and give legally supportable reasons for their conclusions.

55. Conclusion

Civil Law and UAE AI-Generated Evidence Admissibility Standards represents an emerging area in which existing electronic-evidence principles must be applied to technologies capable of creating extremely realistic synthetic material.

The UAE Evidence Law already provides a strong statutory foundation because it expressly recognizes electronic evidence and gives it defined evidentiary treatment. Articles 53–62 are particularly important: they define electronic evidence, identify its forms, regulate its evidentiary value, address challenges to validity, and provide mechanisms for production and verification.

The proper approach is therefore not:

“AI-generated evidence is automatically inadmissible.”

Nor is it:

“AI-generated evidence is automatically reliable.”

The correct approach is:

AI-generated evidence must be evaluated through relevance, authenticity, integrity, attribution, reliability, verification and probative value.

The UAE authorities discussed—particularly Civil Appeal No. 79/2020, Commercial Appeal No. 215/2020, Penal Cassation Nos. 1422/2022, 660/2023, 1093/2019, 1523/2022 and 445/2022—support foundational propositions concerning admissions, expert evidence, judicial evaluation of evidence, logical inference, material defenses and evidentiary weight. They should be treated as analogical authorities, because reported UAE jurisprudence directly addressing modern generative-AI evidence and deepfakes remains comparatively limited.

Ultimately:

AI can generate evidence, reconstruct information and assist forensic analysis—but the UAE court remains responsible for deciding whether the material is authentic, reliable and sufficiently probative to establish the disputed fact.