Civil Law and UAE Machine-Assisted Legal Reasoning Limits

1. Introduction

Machine-assisted legal reasoning means the use of artificial intelligence, machine-learning systems, large language models, automated legal-research tools, predictive analytics, decision-support systems, or other computational technologies to assist lawyers, judges, experts, arbitrators, or litigants in analysing legal issues.

The important legal distinction is:

AI may assist legal reasoning, but assistance does not automatically transfer legal judgment or legal responsibility from the human decision-maker to the machine.

This distinction is particularly important in the UAE because the legal system is increasingly accommodating electronic transactions, digital evidence, AI-related disputes and specialised digital-economy adjudication, while the basic allocation of legal rights and obligations remains centred on legally recognised persons and institutions.

The current UAE position should therefore be understood as machine assistance rather than machine substitution.

2. Current UAE Legislative Framework

A. New UAE Civil Transactions Law

The new Civil Transactions Law, Federal Decree by Law No. 25 of 2025, is now the relevant general civil-law framework. It adopts a hierarchy under which statutory provisions govern matters they address; where legislation is absent, the court moves to Sharia principles, applicable custom, and ultimately principles of natural law and justice. Article 1 also states that there is no room for ijtihad where the legislative text is definitive.

This has an important implication for machine-assisted legal reasoning:

AI cannot create a legal rule merely because an algorithm predicts that a particular result is desirable or common.

The machine must operate within the applicable hierarchy of legal sources.

Example

If an AI system suggests:

“Courts generally allow this contractual term.”

that proposition cannot replace examination of:

1. the applicable statutory provision;
2. mandatory rules;
3. relevant judicial interpretation;
4. contractual terms;
5. factual evidence; and
6. applicable public-order considerations.

3. Electronic Transactions Law and Automated Reasoning

Federal Decree-Law No. 46 of 2021 on Electronic Transactions and Trust Services is especially important.

Article 10 provides that electronic offer and acceptance can form contracts and that a contract does not lose validity merely because it is in electronic-document form.

More importantly, Article 11 expressly recognises contracts made between automated electronic mediums consisting of electronic information systems programmed in advance. Such contracts can be valid, enforceable and legally effective.

This is significant but must not be misunderstood.

Article 11 does NOT mean:

• AI has legal personality;
• AI becomes a judge;
• AI owns property independently;
• AI becomes a lawyer;
• AI possesses independent legal capacity;
• AI can determine the law autonomously.

Instead, it provides functional recognition of automated contracting.

Thus:

Automated execution ≠ autonomous legal personality.

4. What Are the Main Limits of Machine-Assisted Legal Reasoning?

4.1 Limit One — AI Cannot Replace the Applicable Legal Source

A machine may identify:

• statutes;
• cases;
• contractual clauses;
• regulations;
• legal principles;
• patterns in previous judgments.

But it cannot simply replace the legal hierarchy established by legislation.

A prediction such as:

“There is an 82% probability that the claimant will succeed”

is not itself a legal rule.

The court must determine the applicable law and apply it to established facts.

5. Limit Two — AI Hallucination

One of the clearest practical dangers is hallucinated legal authority.

An LLM may generate:

• non-existent cases;
• incorrect case names;
• incorrect article numbers;
• fictional quotations;
• outdated legislation;
• incorrectly stated holdings.

The DIFC Courts have expressly recognised these risks.

Its Practical Guidance Note No. 2 of 2023 states that AI-generated material should be verified and warns about misleading or incorrect information, confidentiality, intellectual-property and data-protection risks. It also expects transparency concerning AI-generated content and says courts may reject AI-generated material under the applicable rules.

Therefore:

Machine-generated legal reasoning requires independent human verification.

6. Limit Three — AI Cannot Become the Source of Truth

Legal reasoning depends upon evidence.

AI can organise evidence but ordinarily cannot determine factual truth merely through statistical inference.

For example, an AI system could compare:

• 10,000 previous contracts;
• thousands of judicial decisions;
• transaction records;
• emails;
• expert reports.

But the court still needs to determine:

• which documents are authentic;
• whether evidence is admissible;
• whether witnesses are credible;
• whether a contractual term applies;
• whether causation is established.

This is particularly important in civil litigation because legal conclusions often depend on fact-sensitive evaluation rather than textual similarity.

7. Limit Four — Explainability

A sophisticated AI system may produce a correct-looking legal conclusion without providing a sufficiently understandable explanation of:

• which evidence was relied upon;
• which legal rule was selected;
• which assumptions were made;
• how conflicting evidence was treated;
• why one interpretation was preferred.

This creates an explainability problem.

A legal judgment ordinarily requires reasons capable of being understood and challenged.

Therefore:

Black-box reasoning is fundamentally problematic where the legal system requires reasoned judicial decision-making.

8. Limit Five — Human Accountability

If a lawyer submits an AI-generated pleading containing false authorities, the lawyer cannot simply say:

“The AI generated it.”

The DIFC Courts' experience demonstrates this point particularly clearly.

In Klesta Eshja & Hair Creators Salon LLC v Salah Masri & Others, CFI 066/2024, the defendants' amended defences had been prepared substantially with AI assistance and contained false references and misleading material. The Court ordered the pleadings struck out and imposed consequential costs orders. A later July 2026 order confirmed that the strike-out was connected to the improper use of AI.

Principle

Human responsibility remains attached to material submitted to the court.

AI assistance does not provide an automatic defence against procedural or professional consequences.

9. Limit Six — AI Cannot Independently Exercise Judicial Discretion

Many civil-law questions involve discretion.

For example:

• whether conduct constitutes abuse of rights;
• whether contractual relief should be granted;
• whether damages are sufficiently proved;
• whether an injunction is appropriate;
• whether evidence is reliable;
• whether mitigation was reasonable;
• whether particular conduct violates public order.

These questions involve contextual judgment.

An AI model can provide analytical assistance, but the legal decision must remain attributable to the legally authorised decision-maker.

10. Limit Seven — Bias in Training Data

Machine reasoning may reproduce biases contained in:

• historical judgments;
• datasets;
• legal databases;
• human annotations;
• transaction records;
• linguistic patterns.

A system trained predominantly on particular jurisdictions may also incorrectly transfer principles from:

• English law;
• US law;
• DIFC law;
• ADGM law;

into a UAE mainland dispute.

This is particularly important because DIFC and ADGM legal systems are not simply interchangeable with mainland UAE civil law.

11. Limit Eight — Jurisdictional Confusion

AI systems may combine authorities from different legal systems.

For example, an AI-generated memorandum might combine:

• UAE Federal legislation;
• DIFC legislation;
• ADGM regulations;
• English common law;
• US precedents.

That can produce a superficially persuasive but legally incorrect argument.

A lawyer must therefore identify:

1. the court;
2. jurisdiction;
3. governing law;
4. contractual choice-of-law clause;
5. applicable legislation;
6. binding or persuasive authority.

12. Limit Nine — Machine Reasoning Does Not Create Legal Personality

The UAE's electronic-transactions framework recognises automated electronic contracting, but that is different from granting AI legal personality.

The important conceptual distinction is:

Current UAE legislation supports the first two and, in specific circumstances, recognises the legal effectiveness of automated transactions.

It does not establish a general category of AI judges or AI legal persons.

13. Important UAE/DIFC Case Laws

Because reported UAE mainland decisions specifically addressing AI-generated legal reasoning remain limited, the following cases combine direct AI authorities with closely related digital-technology and automated-system authorities. DIFC decisions should not be treated as automatically binding precedent for mainland UAE courts.

Case 1 — Klesta Eshja & Hair Creators Salon LLC v Salah Masri & Others

CFI 066/2024, DIFC Court of First Instance

This is one of the most directly relevant UAE-related authorities concerning AI-assisted legal reasoning.

The defendants' amended defences were substantially prepared using AI. The pleadings contained false references and misleading material. The Court struck out the pleadings and made consequential costs orders. The March 2026 reasons expressly recorded the AI-related problem.

Principle

AI-generated legal material must be independently checked.

Importance

The case demonstrates that:

AI assistance does not transfer responsibility for pleadings from the litigant or lawyer to the machine.

14. Case 2 — Stelian Gheorghe v BSA Ahmad Bin Hezeem & Associates LLP & Another

CFI 045/2025, DIFC Court of First Instance

This proceeding involved allegations concerning AI-generated or AI-assisted legal material and issues concerning the reliability of material placed before the court.

The case is relevant to the broader proposition that courts remain concerned with the reliability and legal significance of material generated or influenced by AI.

The DIFC Court's records show subsequent procedural orders in 2026.

Principle

AI-generated legal analysis cannot automatically be treated as authoritative evidence or legal reasoning.

Relevance

The case illustrates the difference between:

AI producing information

and

a court accepting that information as legally reliable.

15. Case 3 — Graciela Limited v Giacobbe

[2014] DIFC CFI 027

This case concerned deliberate interference with the proper functioning of an IT system and a claim involving wrongful interference with property.

The DIFC Court treated the technological system as the subject of legally protected interests rather than as an independent legal person.

Principle

Technology can be:

• property;
• evidence;
• an instrument;
• an object of legal interference;

without itself becoming the legal subject responsible for the consequences.

Importance for machine reasoning

An AI system may participate in an event without becoming the legal bearer of the resulting rights and liabilities.

16. Case 4 — Linux v Lizeth

[2022] DIFC SCT 237

The dispute arose from a Software Development Agreement between two companies concerning software development and alleged contractual breach.

The DIFC Small Claims Tribunal treated the dispute as one between legally recognised contracting parties concerning contractual obligations relating to software.

Principle

Software can be the subject matter of contractual rights and obligations without itself becoming a contracting legal person.

Relevance

The same conceptual distinction is important for AI.

An AI model can be:

• licensed;
• supplied;
• developed;
• modified;
• integrated;
• used;

without automatically becoming the legal party to the underlying contract.

17. Case 5 — ICICI Bank Ltd v Bavaguthu Raghuram Shetty

[2022] DIFC CFI 034

The DIFC Court ultimately entered judgment for ICICI Bank against Mr Shetty in February 2025, with a subsequent unsuccessful renewed application for permission to appeal.

The case is relevant to machine-assisted reasoning because it illustrates the importance of structured human adjudication of a complex financial dispute.

The proceedings involved substantial evidence, expert material and legal submissions rather than automated determination of the dispute.

Principle

Complexity of evidence does not justify replacing legal adjudication with algorithmic prediction.

Application

An AI system can help organise:

• financial transactions;
• documents;
• expert evidence;
• chronology;
• contractual obligations.

But the ultimate legal determination remains a judicial function.

18. Case 6 — Techteryx Ltd v Aria Commodities DMCC & Others

[2025] DIFC DEC 001

This is particularly significant because it was handled by the DIFC Digital Economy Court.

The dispute involved digital assets and approximately USD 456 million in stablecoin-related reserves. The Court granted proprietary and worldwide freezing relief concerning the relevant assets.

The Digital Economy Court is specifically designed to handle sophisticated disputes involving areas such as blockchain, AI, fintech, big data, cloud services and other emerging technologies.

Principle

The legal system can adapt its judicial procedures to technologically complex disputes without transferring legal personality or judicial authority to the technology itself.

Significance

Techteryx illustrates:

Digital sophistication can require specialised adjudication without requiring machine adjudication.

19. Case 7 — Alarabi Investments Ltd v Cron AI Ltd

CFI 030/2025, DIFC Court of First Instance

The defendant in this case was Cron AI Ltd, an AI-related corporate entity.

The proceedings concerned default judgment and subsequent applications concerning the judgment.

Principle

The legally recognised party was the company, not the AI technology itself.

This illustrates an important distinction:

AI company ≠ AI legal person.

A corporation developing or operating AI can possess legal personality under applicable company law; that does not mean the AI system operated by the company independently possesses that personality.

20. Case 8 — Naho v Neukirchi

[2024] DIFC SCT 415

This was an appeal-permission proceeding in the DIFC Small Claims Tribunal. The case demonstrates the continuing role of human judicial review and appellate mechanisms in the DIFC system.

Relevance

Even where digital systems are used in modern litigation, legal reasoning remains embedded in:

• judicial decisions;
• appeal rights;
• procedural safeguards;
• human review.

It is therefore useful as an adjacent procedural authority, rather than as a direct AI-personhood case.

21. The Four Levels of Machine-Assisted Legal Reasoning

A useful analytical model is:

Level 1 — Information Retrieval

AI finds:

• statutes;
• cases;
• regulations;
• contractual provisions.

This is generally the least controversial use.

Level 2 — Analytical Assistance

AI:

• compares authorities;
• summarises evidence;
• identifies inconsistencies;
• creates chronologies;
• proposes arguments.

Human verification remains essential.

Level 3 — Predictive Legal Analysis

AI predicts:

• likely outcomes;
• damages;
• litigation risk;
• possible interpretations.

This is more problematic because probability is not the same as legal determination.

Level 4 — Autonomous Legal Decision-Making

The machine itself would determine:

• liability;
• rights;
• remedies;
• legal interpretation.

This raises the most serious issues concerning:

• authority;
• accountability;
• procedural fairness;
• explainability;
• appeal;
• bias;
• legal personality.

The present UAE framework is much more compatible with Levels 1–2 than with fully autonomous Level 4 adjudication.

22. Machine Reasoning and Judicial Discretion

Judicial discretion is especially difficult to automate.

Consider a contractual dispute where a party seeks compensation.

An AI may identify:

breach → damage → possible compensation.

But a court may additionally have to determine:

• whether the breach caused the loss;
• whether the claimant mitigated the loss;
• whether the loss was foreseeable;
• whether the evidence is reliable;
• whether contractual limitations apply;
• whether public policy affects enforcement;
• whether equitable considerations are relevant.

These are not simply mathematical operations.

23. Machine-Assisted Reasoning and Natural-Law/Justice Principles

The new Civil Transactions Law expressly provides a hierarchy that ultimately reaches natural law and rules of justice where no applicable legislative, Sharia or customary rule exists.

This is particularly significant for AI.

An algorithm normally works through:

data → rules/patterns → output.

Legal reasoning can instead require:

rule → facts → interpretation → context → competing principles → justice.

Consequently, the legal system cannot simply assume that the statistically most common answer is necessarily the legally appropriate answer.

24. Machine Reasoning and Evidence

AI-generated reasoning creates several evidentiary questions:

A. Source authenticity

Where did the AI obtain the proposition?

B. Data integrity

Has the underlying dataset been altered?

C. Reproducibility

Would another system produce the same result?

D. Explainability

Can the reasoning process be sufficiently understood?

E. Reliability

Has the output been independently verified?

F. Bias

Could the training data systematically distort the result?

These concerns explain why the DIFC guidance requires verification of AI-generated material.

25. Machine-Assisted Reasoning and Legal Professional Responsibility

A lawyer using AI should generally verify:

1. every statutory citation;
2. every case citation;
3. the actual case holding;
4. whether the case remains good law;
5. jurisdiction;
6. procedural history;
7. quotations;
8. factual assumptions;
9. confidentiality implications;
10. personal-data implications.

The basic professional principle is:

AI may perform the research task, but the legal professional remains responsible for the legal work product.

The Klesta case provides a particularly concrete illustration of this proposition.

26. Machine-Assisted Legal Reasoning vs Machine Adjudication

27. Practical UAE Applications

A. Contract Review

AI identifies:

• termination clauses;
• indemnities;
• limitation clauses;
• unusual provisions.

Human lawyers make the final legal assessment.

B. Litigation Research

AI identifies potentially relevant cases.

The lawyer verifies the actual judgments.

C. Evidence Management

AI can classify thousands of documents.

The court or lawyer determines evidentiary relevance.

D. Construction Disputes

AI can analyse:

• delay records;
• project schedules;
• payment certificates;
• correspondence.

Experts and courts determine legal causation and contractual responsibility.

E. Digital-Asset Litigation

AI can trace transaction patterns.

Courts determine ownership, tracing and remedies.

Techteryx demonstrates how sophisticated digital-asset disputes can be dealt with through specialised judicial structures.

28. Why the UAE's Digital Economy Court Is Important

The DIFC's Digital Economy Court demonstrates an alternative to simply delegating legal reasoning to machines.

Instead of saying:

“AI will decide technology disputes,”

the institutional model is essentially:

“Human courts develop specialised procedures and expertise for technology disputes.”

The DIFC created its Digital Economy Court Division to address disputes involving technologies including AI, blockchain, big data, fintech, cloud services, robotics and related technologies.

This is an important institutional distinction.

29. Central Legal Problems for the Future

1. Attribution

Who is responsible for an AI-generated legal error?

2. Explainability

How much explanation must an AI system provide?

3. Bias

How should algorithmic bias be detected and challenged?

4. Human oversight

When must a human review an AI recommendation?

5. Confidentiality

Can confidential litigation material be placed into an AI system?

6. Data protection

How can personal information used for machine reasoning be protected?

7. Evidentiary status

When should AI output be accepted as evidence?

8. Professional responsibility

Who is responsible for inaccurate AI-generated legal research?

9. Judicial accountability

Can an AI system ever legitimately exercise judicial discretion?

10. Appeals

How would a litigant challenge an algorithmic legal conclusion?

30. Important Doctrinal Formula

The subject can be reduced to five propositions:

Machine Assistance ≠ Machine Authority

Automation ≠ Legal Personality

Prediction ≠ Legal Judgment

Statistical Probability ≠ Proof

AI Output ≠ Verified Legal Authority

These propositions are particularly useful for examination and legal research.

31. Mainland UAE vs DIFC/ADGM

The DIFC authorities cited above are therefore persuasive/illustrative for UAE technology-law analysis, but they should not be presented as automatically binding precedent for an onshore UAE court.

32. Critical Legal Analysis

The central problem is not whether machines can reason in a computational sense.

Modern AI can:

• identify patterns;
• compare legal texts;
• generate arguments;
• classify evidence;
• predict outcomes;
• summarise authorities.

The legal question is different:

Who has the legal authority to transform that computational output into a binding legal conclusion?

Under the current UAE framework, that authority remains attached to recognised legal institutions and persons.

The law may recognise the effect of an automated transaction, but this does not mean that the machine itself becomes a legal subject.

Similarly, a court may use technology to improve judicial efficiency without delegating the ultimate judicial function to an algorithm.

33. Exam-Oriented Revision Points

Meaning

Machine-assisted legal reasoning is the use of AI and computational systems to support legal research, interpretation, evidence analysis and decision preparation.

Main limits

1. AI cannot replace statutory law.
2. AI cannot automatically establish factual truth.
3. AI-generated authorities must be verified.
4. AI predictions are not legal judgments.
5. Black-box reasoning raises explainability problems.
6. Human accountability remains essential.
7. AI does not automatically acquire legal personality.
8. DIFC/ADGM authorities must be distinguished from mainland UAE law.
9. Digital evidence requires authenticity and reliability.
10. Judicial discretion remains a significant barrier to complete automation.

Key legislation

• Federal Decree by Law No. 25 of 2025 — Civil Transactions Law.
• Federal Decree-Law No. 46 of 2021 — Electronic Transactions and Trust Services.
• UAE Evidence legislation.
• Applicable Civil Procedure legislation.
• DIFC Courts Rules and DIFC AI guidance where DIFC proceedings are involved.

34. Case-Law Revision Table

35. Conclusion

Machine-assisted legal reasoning in the UAE is best understood as a support mechanism rather than an independent source of legal authority.

The UAE's electronic-transactions legislation already recognises legally effective automated transactions, while the DIFC has gone further institutionally by establishing a specialised Digital Economy Court and issuing guidance concerning generative AI in proceedings.

At the same time, the recent Klesta Eshja litigation demonstrates the practical limit: AI-generated legal material must be checked by the humans who rely upon it.

Thus, the emerging UAE principle can be stated as:

AI may assist legal reasoning, but legal validity, accountability, procedural fairness and authoritative judgment remain matters for legally recognised human and institutional decision-makers.

The most important distinction for examination purposes is:

Machine intelligence can assist the process of legal reasoning; it does not, merely by producing a sophisticated legal analysis, acquire the legal authority to decide the dispute.