Civil Law and UAE — Foreclosure of Appeals in Automated Adjudication

1. Introduction

“Foreclosure of appeals in automated adjudication” is an emerging analytical concept rather than a single codified UAE legal doctrine. It describes a situation in which an automated or AI-assisted system produces a decision and the practical or legal ability of the affected person to challenge that decision is reduced, blocked, or made ineffective.

The central legal problem is:

Can an automated decision become effectively final without meaningful human review, reasons, notice, opportunity to challenge, and appellate supervision?

Under current UAE jurisprudential developments, there is no established general rule that an AI or algorithmic output itself constitutes a final judicial judgment immune from appeal. Rather, ordinary principles of jurisdiction, procedural fairness, reasoned decision-making, evidence, human responsibility and appellate review remain highly relevant. Recent DIFC and ADGM decisions concerning AI-assisted litigation reinforce the importance of human verification and reasoned adjudication.

2. Meaning of “foreclosure of appeals”

Here, foreclosure means the practical closing-off of an appellate pathway.

It may occur where:

• an algorithm gives a decision without identifying reasons;
• there is no meaningful human review;
• the affected person does not know how the decision was generated;
• the system does not preserve the relevant input data;
• an appeal is technically available but practically impossible to formulate;
• an automated decision is treated as conclusive without judicial scrutiny;
• procedural deadlines begin without adequate notice;
• the appellate court cannot reconstruct the reasoning.

Core formula

AUTOMATED OUTPUT → NO EXPLANATION → NO EFFECTIVE CHALLENGE → NO MEANINGFUL APPEAL → PRACTICAL FINALITY

3. Important UAE qualification

The UAE must be divided into different legal systems.

Onshore UAE

Federal and emirate courts operate under UAE legislation and established appellate structures.

DIFC

The DIFC Courts have their own procedural and appellate framework.

ADGM

ADGM Courts operate under their own statutory and common-law framework.

Therefore:

A DIFC or ADGM AI-related judgment should not automatically be described as an onshore UAE precedent.

The cases below are principally DIFC and ADGM authorities used by analogy to explain the emerging legal problem.

4. Why appellate review matters in automated adjudication

Appeal serves several functions:

1. correction of legal error;
2. correction of factual error;
3. correction of procedural unfairness;
4. supervision of first-instance reasoning;
5. development of legal precedent;
6. protection of the right of defence;
7. institutional accountability.

If an algorithm produces a conclusion but cannot explain:

• what data it used;
• what legal rule was applied;
• what assumptions were made;
• why one conclusion was preferred;

then an appellate court may have difficulty determining whether the decision was legally correct.

Memory Trigger

No intelligible reasoning → impaired appellate review

5. Oheo Bank v Parker — the leading modern analogy

Oheo Bank v Parker [2025] DIFC CA 006

This is one of the strongest authorities for the proposition that adequate reasons are essential to meaningful appellate review.

The DIFC Court of Appeal stated that failure to provide adequate reasons can itself constitute a ground of appeal. It emphasised that reasons are a function of due process and an essential precondition for satisfactory appellate operation. The appellate court must be able to determine whether the first-instance judge addressed the decisive issues and reasoned logically and rationally.

The Court ultimately set aside the first-instance judgment after finding that the reasons did not provide a satisfactory basis for appellate review.

Application to automated adjudication

If a human judgment must provide enough reasoning for appellate review, an automated decision that provides only:

“Risk score: 92 — application rejected”

would raise an obvious legal-reasoning problem if that output were being treated as the legally determinative decision.

The issue is not that algorithms must necessarily disclose every line of source code.

The issue is whether the legally decisive reasoning can be understood and challenged.

Memory Trigger

Reasons → Review → Appeal

6. Automated adjudication cannot simply eliminate judicial responsibility

An algorithm can:

• classify;
• calculate;
• search;
• predict;
• recommend;
• rank;
• identify patterns.

But these technological functions do not automatically confer legal adjudicative authority.

A legally valid judgment normally depends upon:

• lawful jurisdiction;
• authorised decision-maker;
• applicable law;
• evidence;
• procedural fairness;
• reasoning;
• operative order;
• appeal/review mechanisms.

Formula

TECHNOLOGY ≠ JUDICIAL AUTHORITY

7. Arabyads Holding v Gulrez Alam

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

This ADGM case is extremely important for understanding human responsibility surrounding AI-generated legal material.

Legal representatives used AI-assisted research that resulted in fictitious or inaccurate authorities being included in litigation material. The ADGM Court held that lawyers using AI remain professionally responsible for verifying the accuracy of their research and ordered the relevant legal representatives to pay AED 282,508 in wasted costs on an indemnity basis.

Relevance to automated adjudication

The case does not concern an AI judge.

Its significance is broader:

AI assistance does not transfer legal responsibility from the human professional to the machine.

That principle is highly relevant to automated adjudication.

If an AI system recommends a judicial outcome:

AI recommendation → human judge → verification → legal reasoning → judgment

is materially different from:

AI recommendation → automatic final judgment → no human review

Memory Trigger

AI assists; humans remain accountable.

8. Klesta Eshja — AI material and procedural reliability

Klesta Eshja & Hair Creators Salon LLC v Salah Masri & Others [2024] DIFC CFI 066

This litigation involved substantial procedural applications and the use of litigation materials in circumstances where accuracy and proper procedural handling were important. The DIFC Court continued to exercise judicial control over pleadings, amendments and procedural applications.

Relevance

The important principle for automated adjudication is that technology does not displace procedural control.

A court must still determine:

• what material is admissible;
• what material is relevant;
• whether parties have had an opportunity to respond;
• whether procedural fairness has been maintained.

Memory Trigger

Automated process ≠ uncontrolled process

9. Khaled Al Mheiri v John Cameron

Khaled Salem Musabeh Humad Al Mheiri v John Cameron [2025] DIFC CA 008

The DIFC Court of Appeal strongly emphasised adequate reasoning.

It explained that adequate reasons facilitate:

• appellate review;
• understanding by the losing party;
• identification of factual findings;
• understanding of legal principles;
• identification of the reasoning process.

The Court also stressed that requiring proper reasons helps reduce the risk of inadvertent judicial error.

Automated adjudication application

An algorithmic system should therefore not be allowed to create a situation where:

decision exists → reasoning unavailable → appeal becomes speculative.

The legal system needs enough information to reconstruct the decisive reasoning.

Memory Trigger

Reasoned decision = reviewable decision

10. Pembroke v Paloma — permission to appeal

Pembroke v Paloma [2025] DIFC SCT 946

The DIFC Court of Appeal explained that permission to appeal requires a real prospect of success or another compelling reason for the appeal to be heard.

It also emphasised that a procedural irregularity must be connected to material unfairness or a possible effect on the outcome.

Automated adjudication relevance

Suppose an automated system makes an error.

The affected party must still demonstrate:

• what the error was;
• why it was legally relevant;
• how it affected the decision;
• why appellate intervention is justified.

Therefore, automated adjudication can make the evidentiary burden of challenging a decision particularly important.

Memory Trigger

Appeal requires identifiable error + legal significance

11. Architeriors v Emirates National Investment

Architeriors Interior Design LLC v Emirates National Investment Co LLC [2025/2026] DIFC TCD

The DIFC Technology and Construction Division addressed permission-to-appeal reasoning and emphasised that when a renewed application for permission is refused, the reasons should adequately explain why permission was refused because the refusal is final and conclusive as to that appeal route.

Relevance to automated adjudication

This illustrates an important distinction:

Ordinary decision

May remain subject to appeal.

Final refusal of permission

Can close a particular appellate pathway.

Therefore, when automated decision systems are incorporated into procedures involving permission to appeal, transparency becomes even more important.

Memory Trigger

Closed appellate gateway → stronger need for intelligible reasons

12. Mydlarz v Sadapay Technologies

Eli Mydlarz v Sadapay Technologies Ltd [2025/2026] DIFC

This recent litigation is particularly interesting because it involves a technology company and the DIFC appellate permission process.

The Court of Appeal dismissed Sadapay's renewed application for permission to appeal in January 2026. The order demonstrates that access to appellate review operates through defined procedural rules rather than as an unrestricted second hearing.

Significance

Technology companies remain subject to:

• jurisdictional rules;
• procedural rules;
• permission-to-appeal requirements;
• judicial reasoning;
• costs consequences.

The fact that a company operates a digital platform does not create a special immunity from ordinary procedural law.

Memory Trigger

Digital defendant ≠ special appellate regime

13. Brookfield Multiplex — expert systems do not replace the court

Brookfield Multiplex Constructions LLC v DIFC Investments LLC [2016] DIFC CFI 020

Brookfield concerned a complex construction dispute and the interaction between court proceedings, expert evidence and arbitration.

The case is particularly useful for the principle that technical or expert assistance does not transfer ultimate legal decision-making from the court or tribunal.

Application to AI

An AI system may be more sophisticated than an ordinary expert report.

But sophistication does not itself confer adjudicative authority.

Thus:

AI analysis → evidence/assistance

does not automatically become:

AI analysis → binding legal judgment

Memory Trigger

Technical expertise assists; legal authority decides.

14. Thamer Abdulaziz Albulaihid v Nasser Shehata

Thamer Abdulaziz Albulaihid v Nasser Shehata & Others [2023] DIFC CFI 079

This litigation involved complex factual, technical and evidentiary issues, including questions concerning the respective legal and evidential burdens.

The Court's procedural decisions demonstrate the continuing importance of:

• jurisdiction;
• evidence;
• burden of proof;
• procedural applications;
• opportunity to respond. 

Automated adjudication relevance

An automated system should not be permitted to obscure:

Who bears the legal burden of proving the relevant fact?

AI may process evidence, but it should not silently alter the legal burden of proof.

Memory Trigger

Algorithmic confidence ≠ legal burden

15. Automated decision versus automated judgment

This distinction is essential.

Automated administrative decision

Example:

• automated benefit calculation;
• automated fraud flag;
• automated licence screening;
• automated credit assessment.

The law may permit automated assistance subject to statutory requirements.

Automated judicial decision

A different problem arises if software itself purports to exercise judicial authority.

The critical questions become:

• Who is legally the judge?
• Who has jurisdiction?
• Who heard the parties?
• Who assessed the evidence?
• Who gave reasons?
• Who bears responsibility?
• Who signs the judgment?
• What appeal is available?

Formula

Automated administrative decision ≠ automated judicial judgment

16. Human review must be meaningful

Simply placing the word “human” somewhere in the process is not enough conceptually.

There is a major difference between:

Rubber-stamp review

AI says:

Reject.

Human says:

I agree.

and:

Substantive review

Human:

1. examines the relevant evidence;
2. identifies the legal issues;
3. considers the algorithmic output;
4. considers contrary material;
5. independently applies the law;
6. gives reasons.

For purposes of meaningful appellate review, the second model provides a much stronger basis.

17. Explainability and appellate review

Explainability has at least five dimensions.

1. Input transparency

What information was supplied?

2. Method transparency

What process produced the result?

3. Output transparency

What exactly did the system conclude?

4. Legal transparency

How did the output relate to the applicable legal rule?

5. Review transparency

How can the affected party challenge the result?

Formula

INPUT → PROCESS → OUTPUT → LEGAL REASON → CHALLENGE

18. Why black-box systems create appellate difficulty

Suppose an AI system decides:

“Claim dismissed.”

But the record does not disclose:

• which documents were relied upon;
• which evidence was rejected;
• which legal rule was applied;
• whether contradictory evidence was considered;
• whether the system made a data error;
• whether the result was generated by a statistical model or legal rule.

An appeal becomes difficult because the appellate court cannot determine why the decision was reached.

This is exactly the type of problem highlighted by Oheo Bank: adequate reasons are necessary for the appellate court to identify the issues considered, evidence relied upon and reasoning process.

19. Foreclosure through procedural opacity

Appeal can be indirectly foreclosed through opacity.

Stage 1

AI produces decision.

Stage 2

No explanation is supplied.

Stage 3

Party cannot identify the error.

Stage 4

Appeal grounds become speculative.

Stage 5

Permission to appeal is refused because no sufficiently identifiable error is demonstrated.

Result

Formal appeal right exists, but effective appeal becomes difficult.

This is practical foreclosure, rather than necessarily formal legal abolition of appeal.

20. Foreclosure through inaccessible evidence

Another problem arises if the algorithm's underlying data are unavailable.

For example:

• proprietary model;
• inaccessible training data;
• deleted logs;
• undisclosed scoring variables;
• third-party vendor system;
• encrypted audit trail.

The affected party may be unable to demonstrate that:

• the data were wrong;
• the model was biased;
• the calculation was defective;
• relevant evidence was ignored.

Principle

An appeal cannot be fully effective if the record necessary to test the decision does not exist or cannot be examined.

21. AI-generated reasons

There is another difficult question:

Can a court rely upon AI-generated reasons?

The critical distinction is between:

AI drafting assistance

AI helps organise or draft reasons.

versus

AI determining the legal reasoning

AI itself determines:

• facts;
• legal rules;
• credibility;
• causation;
• liability;
• remedy.

The first can potentially be treated as a technological drafting tool subject to human verification.

The second raises much more fundamental questions concerning judicial authority and responsibility.

The reasoning in Arabyads strongly supports the proposition that AI-generated legal material must be verified by the responsible human professional.

22. AI hallucinations and appellate foreclosure

Imagine an AI adjudication system cites:

Case A [2021]

but Case A does not exist.

The affected party may not immediately know that.

If the decision is then treated as final, the party faces two problems:

1. substantive error;
2. procedural inability to expose the error.

Arabyads demonstrates the first-order principle that AI-generated legal material requires human verification. The ADGM Court treated failure to verify AI-assisted legal research as sufficiently serious to justify substantial wasted-cost consequences.

Memory Trigger

Hallucinated authority + no verification = procedural risk

23. Automated fact-finding

AI may be used to:

• classify documents;
• detect inconsistencies;
• identify relevant evidence;
• summarise testimony;
• compare transactions;
• analyse financial records.

But automated fact-finding must be distinguished from judicial fact-finding.

A model can identify:

“These two witness statements contain different dates.”

The court must still decide:

“Which evidence is credible and what legal significance does the discrepancy have?”

Formula

AI detects → Human evaluates → Court decides

24. Automated credibility assessment

This is particularly sensitive.

Suppose AI gives a witness a:

“Credibility score = 0.23”

That score cannot automatically replace judicial evaluation.

Credibility involves:

• evidence;
• consistency;
• context;
• cross-examination;
• documentary corroboration;
• legal burden;
• judicial assessment.

An algorithmic probability is not automatically a legal finding.

25. Appeal and the right of defence

Meaningful appellate review is closely connected to the right of defence.

A party needs to know:

• the case against it;
• evidence relied upon;
• legal basis;
• factual findings;
• reasons;
• available challenge mechanism.

If an algorithm secretly determines the decisive issue, the party may be unable to meaningfully defend itself.

This is particularly problematic where the automated result affects:

• property;
• money;
• contractual rights;
• financial accounts;
• regulatory status;
• business licences;
• civil liability.

26. Algorithmic finality versus judicial finality

These should never be confused.

Algorithmic finality

The software stops changing its output.

Judicial finality

The legally authorised judicial process has reached a stage at which the decision is no longer subject to an available ordinary appeal or has otherwise become final under applicable law.

Therefore:

A frozen algorithmic output is not necessarily a legally final judgment.

Memory Trigger

Software finality ≠ legal finality

27. Appealability of an algorithmic recommendation

Suppose:

AI recommends dismissal → Judge independently reviews → Judge issues reasoned judgment.

The appeal normally concerns the judgment, not merely the algorithmic recommendation.

But if:

AI determines dismissal → human merely signs → judgment adopts output without independent reasoning,

the algorithmic process may become central to a challenge based on:

• procedural irregularity;
• inadequate reasons;
• failure to consider evidence;
• improper delegation;
• denial of meaningful participation.

28. AI and appellate record preservation

A reliable automated adjudication architecture should preserve:

• input data;
• documents considered;
• model/version;
• relevant prompts or instructions;
• decision tree;
• algorithmic output;
• confidence levels where relevant;
• human interventions;
• corrections;
• final judicial reasoning.

Formula

AUDIT TRAIL = INPUT + PROCESS + OUTPUT + HUMAN REVIEW + FINAL REASONING

Without such a record, appellate review may become practically impossible.

29. Vendor responsibility

Suppose a court uses an external AI vendor.

Potential responsibility questions include:

• Who owns the model?
• Who controls the data?
• Who validates the model?
• Who maintains it?
• Who corrects errors?
• Who preserves logs?
• Who is responsible for cybersecurity?
• Can the court independently review the system?

A vendor's contractual confidentiality cannot automatically override the requirements of lawful adjudication.

30. Trade secrets versus appellate transparency

AI vendors may argue:

“Our algorithm is proprietary.”

That may justify protection of genuine trade secrets in some circumstances.

But it does not automatically answer whether the affected party is entitled to enough information to challenge a legally consequential decision.

The balance may therefore be:

TRADE SECRET PROTECTION

versus

PROCEDURAL FAIRNESS + EFFECTIVE REVIEW

The solution may involve:

• confidential disclosure;
• expert inspection;
• redaction;
• controlled access;
• independent technical examination.

31. Automated adjudication and evidence

Evidence should be analysed separately from the algorithm.

Question 1

Is the underlying data authentic?

Question 2

Is it complete?

Question 3

Was it altered?

Question 4

Was the algorithm correctly applied?

Question 5

Was contradictory evidence considered?

Question 6

What weight should the output receive?

Formula

DATA AUTHENTICITY → MODEL RELIABILITY → OUTPUT → HUMAN EVALUATION

32. Expert evidence and automated systems

Technical experts can explain:

• architecture;
• model operation;
• source data;
• system error;
• statistical methodology;
• cybersecurity;
• algorithmic bias.

But the expert does not decide the legal question.

This is consistent with the reasoning reflected in Brookfield Multiplex concerning the role of technical expertise within judicial adjudication.

Memory Trigger

Expert explains the system; court determines its legal consequence.

33. Permission-to-appeal problem

Automated adjudication creates an especially difficult problem where the appellate system requires permission.

The party may have to show:

• a real prospect of success;
• legal error;
• serious procedural irregularity;
• another compelling reason.

That is difficult if the underlying algorithm is opaque.

Thus:

Opaque AI → difficult error identification → difficult permission application

This is why the quality of first-instance reasons matters so much.

The DIFC appellate authorities repeatedly connect adequate reasons with the effective operation of appeal.

34. Six major case-law authorities

The cases do not establish a single UAE doctrine declaring all automated adjudication unlawful. Rather, they provide principles from which the legal risks of opaque or effectively unreviewable automated decision-making can be analysed.

35. Distinctions to remember

1. Automated decision ≠ automated judgment

A software decision may be administrative or advisory; a judgment requires legal authority.

2. AI assistance ≠ AI adjudication

Using AI to assist a judge is different from delegating adjudicative authority to AI.

3. Technical opacity ≠ automatic illegality

Opacity becomes legally significant when it prevents meaningful challenge or review.

4. Appeal right ≠ effective appeal

A formal appeal route may exist while practical review is impaired by lack of reasons or evidence.

5. AI output ≠ evidence of truth

Algorithmic output must itself be assessed.

6. AI confidence ≠ legal proof

A probability score cannot automatically replace the applicable evidentiary standard.

7. Algorithmic finality ≠ legal finality

Software stopping its calculation does not create judicial finality.

8. Human presence ≠ meaningful human review

A person merely approving an automated result may not provide substantive independent review.

36. Advanced Master Formula

AUTOMATED ADJUDICATION

↓

LEGAL AUTHORITY

↓

JURISDICTION

↓

INPUT DATA

↓

EVIDENCE

↓

ALGORITHMIC PROCESS

↓

AI OUTPUT

↓

HUMAN VERIFICATION

↓

LEGAL REASONING

↓

REASONS FOR DECISION

↓

NOTICE TO PARTIES

↓

CHALLENGE

↓

PERMISSION / APPEAL

↓

APPELLATE REVIEW

↓

FINALITY

37. Anti-Foreclosure Safeguards

A legally robust automated adjudication system should ideally contain:

1. Human judicial responsibility
2. Traceable evidence
3. Preserved audit trail
4. Identifiable legal rules
5. Reasoned decision
6. Disclosure of material AI involvement
7. Opportunity to challenge
8. Access to relevant underlying material
9. Independent human review
10. Appeal mechanism
11. Correction procedure
12. Record preservation

Formula

TRANSPARENCY + HUMAN CONTROL + REASONS + CHALLENGE + APPEAL = EFFECTIVE REVIEWABILITY

38. Ultra-Fast Exam Revision

1. AI output is not automatically a judgment.
2. Legal authority must come from law, not software.
3. Human responsibility remains important.
4. Reasons are essential for appellate review.
5. Opaque reasoning can impair appeal.
6. Formal appeal ≠ effective appeal.
7. Algorithmic finality ≠ legal finality.
8. AI-generated authorities require verification.
9. AI evidence must be authenticated and evaluated.
10. Algorithmic probability ≠ legal proof.
11. Expert explanation ≠ judicial determination.
12. The legal burden remains a legal question.
13. Procedural irregularity must be connected to material unfairness.
14. Permission-to-appeal rules remain relevant.
15. Technology companies remain subject to ordinary procedural rules.
16. A human merely pressing “approve” is not necessarily meaningful review.
17. Audit trails are important for appellate reconstruction.
18. Trade secrecy does not automatically eliminate procedural fairness.
19. DIFC/ADGM authorities are distinct from onshore UAE precedent.
20. Effective judicial review requires an intelligible decision record.

39. Final Principle

UAE Automated Adjudication and Appeal =

LAWFUL AUTHORITY + HUMAN RESPONSIBILITY + EVIDENCE + EXPLAINABILITY + REASONS + CHALLENGE + APPELLATE REVIEW + FINALITY

The most important emerging principle is:

Automation should not transform a legally reviewable decision into an effectively unreviewable one merely because the reasoning was produced by software.

The recent Oheo Bank judgment provides the clearest appellate analogy: adequate reasons are not merely a matter of presentation; they are part of due process and are necessary for the appellate court to perform its review function.

And Arabyads supplies the complementary AI principle: using AI does not transfer responsibility away from the human legal actor.

Final Memory Line

“Software may generate an answer; lawful adjudication requires authority, reasons, human responsibility, challenge and review.”