AI and Judicial Review

Artificial Intelligence is increasingly being used in public administration for data analysis, risk assessment, fraud detection, welfare administration, regulatory monitoring and decision support. As government authorities adopt AI-assisted systems, a new question arises in Administrative Law: how should courts review an administrative decision that has been partly or substantially generated by an AI system?

Judicial review traditionally examines whether a public authority has acted legally, fairly, rationally and within the limits of its power. The use of AI does not remove these requirements. Instead, it creates additional questions concerning algorithmic reasoning, data accuracy, transparency, bias, human oversight and accountability.

AI therefore changes the method through which administrative decisions may be reached, but it does not fundamentally change the principle that the exercise of public power remains subject to law.

Meaning of AI and Judicial Review

AI and judicial review refers to the application of traditional judicial-review principles to administrative decisions that are:

• Made with the assistance of AI;
• Recommended by an AI system;
• Based on algorithmic analysis;
• Generated through automated decision-making; or
• Substantially influenced by machine-learning systems.

The court is not ordinarily required to determine whether an AI system is technologically sophisticated or commercially effective.

The central legal questions remain:

• Was the authority legally empowered to act?
• Was the correct legal standard applied?
• Was the procedure fair?
• Were relevant considerations taken into account?
• Was the decision arbitrary or unreasonable?
• Were Fundamental Rights affected?
• Was the decision proportionate where proportionality applies?
• Can the authority adequately explain and defend the decision?

Why AI Creates New Judicial Review Questions

Traditional administrative decisions are generally capable of being traced to an identifiable decision-maker and stated reasons.

AI-assisted decisions may involve:

Data → Algorithm → Prediction or Recommendation → Official Review → Administrative Decision

This creates additional questions.

For example:

• What data was used?
• Was the data accurate?
• What role did the AI system play?
• Was the algorithm legally authorised?
• Did the official independently consider the matter?
• Was the system biased?
• Can the decision be explained?
• Can the affected person effectively challenge it?

These questions supplement rather than replace traditional grounds of judicial review.

AI Does Not Escape Judicial Review

The use of an algorithm does not make an administrative decision immune from judicial review.

If a government authority uses an AI system to make or influence a decision, the authority remains responsible for exercising its statutory power according to law.

A public authority cannot ordinarily defend an unlawful decision merely by stating that:

“The algorithm made the decision.”

The legal responsibility remains connected to the public authority exercising the statutory power.

Principle of Legality

The first question in judicial review is whether the authority acted within its legal powers.

An AI system cannot independently create administrative authority.

If legislation permits an authority to take action only when specified conditions exist, an AI system must operate within those conditions.

A court may therefore examine:

• The enabling statute;
• Delegated legislation;
• Administrative rules;
• The purpose of the power;
• The criteria applied by the system; and
• The actual decision taken.

If an AI system applies criteria that have no legal basis, the resulting administrative action may be vulnerable to judicial review.

AI and Ultra Vires

The doctrine of ultra vires remains relevant to AI-assisted administration.

An AI-based administrative action may be challenged where:

• The authority had no jurisdiction;
• The system applies criteria beyond the statute;
• A mandatory statutory procedure was ignored;
• A legal power was used for an improper purpose; or
• The authority effectively delegated a statutory discretion to a system without lawful authority.

The use of technology does not alter the basic rule that public authorities cannot exceed their legal powers.

AI and Article 14

Article 14 is particularly significant in reviewing AI-assisted administrative action.

An AI system may produce arbitrary or unequal outcomes because of:

• Biased training data;
• Incomplete datasets;
• Incorrect classifications;
• Proxy variables;
• Historical patterns of discrimination; or
• Improper system design.

E.P. Royappa v. State of Tamil Nadu (1974)

The Supreme Court connected Article 14 with the principle against arbitrary State action.

This principle is relevant to AI because algorithmic administration remains a form of State action.

Maneka Gandhi v. Union of India (1978)

The Supreme Court strengthened the relationship between Articles 14, 19 and 21 and emphasised fairness and non-arbitrariness in governmental action.

AI-assisted administration must therefore remain consistent with these constitutional requirements.

AI and Natural Justice

Natural justice remains relevant when AI contributes to an administrative decision.

Its application depends upon the nature of the decision, the governing law and the consequences for the affected person.

Important questions include:

• Was adequate notice provided?
• Was the person given an opportunity to respond?
• Was relevant adverse material disclosed where required?
• Was the decision made impartially?
• Were legally required reasons provided?

An AI system cannot be used to circumvent a procedural safeguard that the law otherwise requires.

AI and the Right to Be Heard

Suppose an AI system flags a person as ineligible for a government benefit.

If the governing legal framework requires an opportunity to respond before the benefit is denied, the authority cannot necessarily treat the AI output as conclusive.

The affected person may need an opportunity to:

• Challenge the underlying information;
• Explain exceptional circumstances;
• Correct inaccurate data; and
• Contest the proposed administrative action.

This is particularly important where the AI output has significant consequences.

AI and Disclosure of Information

Judicial review may become difficult if an affected person has no meaningful information about the basis of an AI-assisted decision.

Relevant information may include:

• The legal criteria applied;
• The factual information relied upon;
• The role played by the AI system;
• The significant factors influencing the result; and
• The available review mechanism.

This does not necessarily mean that every aspect of an algorithm or its source code must be disclosed in every case.

The appropriate level of disclosure depends upon the legal framework, the nature of the decision and legitimate interests such as security and confidentiality.

AI and Duty to Give Reasons

Reasoned decision-making is an important safeguard in Administrative Law.

S.N. Mukherjee v. Union of India (1990)

The Supreme Court recognised the importance of recording reasons in administrative and quasi-judicial decisions, subject to recognised exceptions.

AI makes the duty to give reasons particularly significant because an unexplained algorithmic output may make it difficult for an affected person to understand or challenge an adverse decision.

A statement such as “the system determined that the applicant was ineligible” may not, by itself, provide an adequate explanation where reasons are legally required.

AI and Explainability

Explainability refers to the ability to understand how an AI system reached or contributed to an outcome.

In judicial review, explainability may be relevant because courts need to determine whether:

• The correct legal criteria were used;
• Relevant factors were considered;
• Irrelevant factors influenced the outcome;
• The system operated consistently;
• The decision was rational; and
• The authority acted within its statutory purpose.

The more significant the impact of the decision on rights or interests, the greater the practical importance of being able to understand its basis.

AI and Relevant and Irrelevant Considerations

One established ground of judicial review concerns consideration of relevant and irrelevant factors.

AI systems may process hundreds or thousands of variables.

The legal question is not simply whether the system considered a large quantity of information.

The question is whether the information considered was legally relevant to the administrative decision.

A system may therefore produce a technically sophisticated but legally defective decision if it relies upon irrelevant factors.

AI and Improper Purpose

Administrative powers must be exercised for the purpose for which they were granted.

An AI system used for one legitimate administrative purpose cannot automatically be used for an unrelated purpose merely because the underlying data is available.

For example, information collected for a specific administrative function should not automatically be repurposed for an unrelated exercise without appropriate legal authority.

Judicial review may therefore examine the purpose for which an AI system and its underlying data are being used.

AI and Fettering of Administrative Discretion

Administrative authorities sometimes possess statutory discretion.

An AI system may create a rigid decision-making structure that effectively eliminates that discretion.

For example, if an official is legally required to consider individual circumstances but automatically follows an AI-generated recommendation in every case, the official may fail to exercise the discretion entrusted by law.

Judicial review can therefore examine whether:

• The official independently considered the matter;
• The AI output was treated as merely advisory or effectively binding;
• Individual circumstances were considered; and
• The statutory discretion was properly exercised.

AI and Wednesbury Unreasonableness

The principle of unreasonableness remains relevant to AI-assisted decisions.

Associated Provincial Picture Houses Ltd. v. Wednesbury Corporation (1948)

The traditional Wednesbury principle concerns administrative decisions that fall outside the range of reasonable decisions open to the authority.

An AI-generated outcome may therefore be challenged if the public authority relies upon an irrational or manifestly unreasonable result without properly examining it.

The fact that a computer generated the result does not itself establish reasonableness.

AI and Proportionality

Proportionality is particularly important where AI-assisted administrative action affects Fundamental Rights.

Modern Dental College & Research Centre v. State of Madhya Pradesh (2016)

The Supreme Court recognised proportionality as an important constitutional method for examining restrictions on rights.

K.S. Puttaswamy v. Union of India (2017)

The constitutional right to privacy has particular relevance to AI systems that collect, analyse or infer information about individuals.

Where AI-based administration interferes with privacy or another protected right, judicial review may consider:

• The legitimate objective;
• The connection between the measure and the objective;
• Whether a less restrictive approach is available where applicable; and
• Whether the impact on rights is justified.

AI and Privacy

AI systems can process large amounts of personal information and may generate inferences from data that were not expressly provided by an individual.

Judicial review may therefore examine:

• The legal basis for data collection;
• The purpose of processing;
• The extent of data collection;
• Safeguards against misuse;
• Data retention;
• Access controls; and
• The proportionality of the interference.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2017)

The Supreme Court recognised privacy as a constitutionally protected right.

This provides an important constitutional framework for reviewing government use of data-intensive AI systems.

AI and Algorithmic Bias

Algorithmic bias can become a ground of legal challenge where it results in unlawful discrimination or arbitrary administrative action.

Bias may originate from:

• Historical government data;
• Under-represented groups;
• Incorrect datasets;
• Proxy variables;
• System design;
• Data labelling; or
• Human assumptions embedded in the model.

The court may therefore need to examine not only the final outcome but also the decision-making process that produced it.

AI and Equality

An algorithm may appear neutral while producing unequal effects.

For example, a system may use a factor that appears unrelated to a protected characteristic but operates as a proxy for it.

The constitutional question is whether the resulting classification or treatment has a sufficient legal and rational basis.

The mere technological neutrality of an algorithm does not automatically establish constitutional validity.

AI and Procedural Impropriety

Procedural impropriety may arise where an AI system causes an authority to depart from a procedure required by law.

Examples include:

• Failure to issue required notice;
• Failure to consider representations;
• Failure to disclose required material;
• Failure to provide a hearing;
• Failure to record required reasons; or
• Failure to follow statutory safeguards.

An AI system must therefore be integrated into the legally prescribed administrative procedure.

AI and Jurisdictional Error

Jurisdictional error occurs when an authority acts:

• Without legal authority;
• Beyond the limits of its jurisdiction; or
• On an incorrect understanding of the scope of its legal power.

AI can contribute to such an error if the system is programmed according to an incorrect interpretation of the governing law.

The responsibility ultimately remains with the public authority using the system.

AI and Error of Law

AI systems may be trained or programmed to apply legal rules.

If the system applies an incorrect legal rule, the resulting administrative decision may contain an error of law.

A court can examine the legal basis of the decision rather than accepting the algorithm’s interpretation merely because it is technologically sophisticated.

AI and Error of Fact

AI systems can also produce factual errors.

Examples include:

• Incorrect identity records;
• Duplicate records;
• Outdated information;
• Incorrect classification;
• Misidentified transactions; or
• Faulty data matching.

The significance of an error of fact depends upon the applicable legal framework and the nature of the decision.

Where an administrative decision depends upon incorrect factual information, judicial review may become relevant, particularly where the error affects jurisdiction, fairness or the legality of the outcome.

AI and Human Oversight

Meaningful human oversight can provide an important safeguard.

An official reviewing an AI recommendation should be able, where appropriate, to:

• Question the result;
• Examine underlying information;
• Consider individual circumstances;
• Correct obvious errors;
• Seek additional evidence; and
• Depart from the AI recommendation when legally justified.

Human oversight becomes particularly important when the decision has serious consequences for an individual’s rights or interests.

AI and Administrative Accountability

A fundamental problem with AI-assisted administration is the possibility of an accountability gap.

Several actors may be involved:

• Government department;
• Administrative officer;
• Technology provider;
• Data provider;
• Software developer; and
• System administrator.

For purposes of public law, technological complexity should not obscure the identity of the public authority responsible for exercising the statutory power.

AI and Judicial Review of Government Contracts

Government may procure AI systems from private companies.

This creates additional questions concerning:

• Tender conditions;
• Procurement fairness;
• Confidentiality;
• Intellectual property;
• Data ownership;
• Algorithmic transparency; and
• Contractual responsibility.

Tata Cellular v. Union of India (1994)

The Court’s principles concerning judicial review of government contracts remain relevant to procurement of AI systems. Courts generally examine legality, fairness and the decision-making process rather than substituting their own commercial judgment.

AI and Public Procurement

AI systems used by government should be procured through legally valid procedures.

Issues may include:

• Equal treatment of bidders;
• Transparent evaluation;
• Technical criteria;
• Security requirements;
• Vendor accountability;
• Data protection;
• Auditability; and
• Contractual safeguards.

A procurement process cannot become immune from judicial review simply because the subject matter is technologically complex.

AI and Welfare Decisions

AI-assisted welfare administration can affect access to:

• Subsidies;
• Pensions;
• Scholarships;
• Food assistance;
• Healthcare benefits; and
• Other government schemes.

Wrongful algorithmic exclusion may have serious consequences.

Judicial review may therefore examine whether:

• The eligibility criteria were legally correct;
• The underlying data was accurate;
• The person was given appropriate procedural protection;
• The authority considered exceptional circumstances; and
• An effective correction or grievance mechanism existed.

AI and Automated Risk Assessment

Regulatory authorities may use AI to identify persons or entities considered high-risk.

Risk scoring may assist in:

• Tax enforcement;
• Financial regulation;
• Environmental compliance;
• Fraud detection; and
• Regulatory inspections.

However, a risk score should not automatically be treated as proof of wrongdoing.

Administrative authorities must distinguish between risk prediction and legally established facts.

AI and Evidence

AI-generated outputs may influence administrative decisions, but their legal significance depends upon the applicable law.

Questions may arise concerning:

• Reliability;
• Accuracy;
• Source data;
• Methodology;
• Auditability;
• Human verification; and
• Opportunity to challenge the information.

An administrative authority should not necessarily treat an algorithmic prediction as equivalent to independently established evidence.

AI and Judicial Deference

Courts may recognise the technical expertise of administrative authorities, particularly in complex regulatory matters.

However, technical complexity does not eliminate judicial review.

The court may distinguish between:

Technical or policy judgment, where administrative expertise may receive appropriate respect, and

Legal questions, such as whether the authority acted within statutory power or violated constitutional requirements.

Judicial review therefore remains concerned with the legality of the exercise of public power.

AI and Separation of Powers

AI can also raise separation-of-powers concerns.

Administrative agencies may use technology to implement legislation.

However, an administrative system should not effectively create new legal rules beyond the authority delegated by the legislature.

The distinction between:

• Legislative policy;
• Administrative implementation; and
• Judicial review

must remain clear.

AI and Constitutional Remedies

Where AI-assisted administrative action violates Fundamental Rights or involves unlawful State action, constitutional remedies may become relevant.

Article 32

Article 32 provides a constitutional remedy for enforcement of Fundamental Rights before the Supreme Court.

Article 226

Article 226 gives High Courts broad writ jurisdiction, including review of administrative action.

Article 227

Article 227 provides supervisory jurisdiction over courts and tribunals within the High Court’s territorial jurisdiction.

The availability of a particular remedy depends upon the nature of the dispute and applicable law.

AI and Writ Jurisdiction

A writ petition may potentially challenge an AI-assisted administrative decision where appropriate grounds exist.

Possible grounds include:

• Lack of jurisdiction;
• Violation of natural justice;
• Arbitrariness;
• Mala fides;
• Improper purpose;
• Violation of Fundamental Rights;
• Relevant and irrelevant considerations;
• Error of law; and
• Proportionality where applicable.

The court may focus on the legality of the administrative action rather than attempting to design or operate the AI system itself.

Remedies in AI-Related Judicial Review

Depending upon the circumstances, courts may provide remedies such as:

• Certiorari to quash an unlawful decision;
• Mandamus to require performance of a public duty;
• Directions for reconsideration;
• Orders requiring procedural safeguards;
• Directions to correct administrative records;
• Constitutional compensation in appropriate cases involving serious rights violations; and
• Other appropriate public-law remedies.

The precise remedy depends upon the facts and governing law.

Important Cases

E.P. Royappa v. State of Tamil Nadu (1974)

Connected Article 14 with protection against arbitrary State action. It provides an important constitutional foundation for examining arbitrary outcomes generated through administrative systems.

Maneka Gandhi v. Union of India (1978)

Emphasised fairness and non-arbitrariness in governmental procedure and strengthened the relationship between Articles 14, 19 and 21.

S.N. Mukherjee v. Union of India (1990)

Recognised the importance of giving reasons in administrative and quasi-judicial decisions, subject to recognised exceptions.

Tata Cellular v. Union of India (1994)

Established important principles governing judicial review of government contracts and procurement decisions.

L. Chandra Kumar v. Union of India (1997)

Reaffirmed the constitutional importance of judicial review under Articles 226/227 and Article 32.

State of Tamil Nadu v. P. Krishnamurthy (2006)

Explained important principles governing judicial review of subordinate legislation.

Modern Dental College & Research Centre v. State of Madhya Pradesh (2016)

Recognised proportionality as an important method of constitutional review where rights are restricted.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2017)

Recognised privacy as a constitutionally protected right and established an important constitutional basis for examining data-intensive government technologies.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2018)

Considered the Aadhaar framework and addressed issues involving digital identity, welfare administration, privacy and proportionality.

Grounds of Judicial Review in the AI Context

Traditional Ground	AI-Related Question
Illegality	Did the AI-assisted decision exceed statutory power?
Jurisdictional error	Did the authority act beyond its legal jurisdiction?
Improper purpose	Was AI or its data used for an unauthorised purpose?
Relevant considerations	Did the system consider legally relevant factors?
Irrelevant considerations	Did irrelevant variables influence the outcome?
Arbitrariness	Did the system produce an unjustified or arbitrary result?
Procedural impropriety	Were required hearings or procedures followed?
Natural justice	Could the affected person meaningfully challenge the decision?
Unreasonableness	Was the authority’s reliance on the AI output unreasonable?
Proportionality	Was the technological interference with rights justified?
Mala fides	Was the technology used for an improper or dishonest purpose?
Error of law	Was the governing legal rule incorrectly understood or applied?

Challenges Before Courts

AI may create practical difficulties for judicial review.

Technical Complexity

Judges may have to understand complicated systems and methodologies.

Black-Box Systems

Some AI models may not provide easily understandable explanations.

Private Technology Providers

Government systems may be developed by private companies, creating issues concerning confidentiality and contractual restrictions.

Data Volume

AI systems may rely upon enormous quantities of information that are difficult to examine manually.

Rapid Technological Change

Judicial principles may need to be applied to technologies that develop faster than legislation.

Institutional Competence

Courts must distinguish legal review from technical or policy choices that fall within administrative expertise.

Principles for Judicial Review of AI

A court examining AI-assisted administrative action may need to consider:

1. Legal authority — Was the use of AI authorised by law?
2. Purpose — Was the system used for a lawful administrative purpose?
3. Data — Was relevant and reasonably reliable information used?
4. Equality — Did the system produce unlawful discrimination or arbitrariness?
5. Procedure — Were applicable procedural safeguards followed?
6. Reasons — Can the administrative authority adequately explain the decision?
7. Human oversight — Did a responsible official properly supervise the process?
8. Privacy — Was personal information processed consistently with applicable law?
9. Proportionality — Where rights are restricted, was the measure legally justified and proportionate?
10. Remedy — Does the affected person have an effective mechanism to challenge the decision?

Key Points for Examination

• AI does not place administrative action outside judicial review.
• The public authority remains responsible for the exercise of its statutory power.
• Legality remains the first requirement.
• Article 14 is relevant to algorithmic arbitrariness and discrimination.
• Article 21 and privacy are important where AI processes personal information.
• Natural justice remains applicable where its requirements are attracted.
• AI-generated outputs do not automatically constitute legally sufficient reasons.
• S.N. Mukherjee is important for the duty to give reasons.
• Puttaswamy (2017) is important for privacy and data-intensive AI systems.
• Algorithmic bias can raise Article 14 concerns.
• Incorrect data can result in unlawful administrative decisions.
• AI should not unlawfully fetter statutory administrative discretion.
• Human oversight is particularly important in high-impact decisions.
• Courts can examine AI-assisted decisions through established grounds such as illegality, irrationality, procedural impropriety and proportionality.
• Technical complexity does not make an administrative decision immune from judicial scrutiny.
• Judicial review generally examines the legality of the decision-making process, rather than replacing the administration’s technical or policy judgment with that of the court.

Conclusion

Artificial Intelligence introduces new technological possibilities into administrative decision-making, but it does not fundamentally alter the constitutional principle that public power is subject to law.

When AI assists or influences an administrative decision, judicial review may require courts to look beyond the final output and examine the legal authority, data, decision-making process, procedural safeguards, reasons, human oversight and effect on constitutional rights.

The major challenge is to ensure that technological complexity does not create an accountability gap. An administrative authority cannot avoid legal responsibility by attributing an adverse decision to an algorithm.

AI and judicial review therefore represent an evolving area of Administrative Law in which traditional principles of legality, natural justice, equality, reasonableness, proportionality, privacy and accountability must be applied to technologically assisted exercises of public power.