Algorithmic Equity Claims in India

1. Meaning and Concept

Algorithmic Equity Claims refer to legal claims arising when an algorithm, artificial-intelligence system, automated decision-making tool, scoring model, or data-driven system produces unequal, discriminatory, arbitrary, exclusionary, or otherwise unfair outcomes.

The expression is not presently a standalone statutory cause of action in Indian law. An affected person normally has to formulate the claim through existing legal doctrines, including:

• Article 14 equality and non-arbitrariness;
• Articles 15 and 16 discrimination protections;
• Article 21 dignity, privacy and informational autonomy;
• disability rights;
• employment and labour law;
• consumer protection;
• contract law;
• negligence;
• administrative law;
• data-protection law;
• sector-specific regulation; and
• judicial review.

The central question is:

Can an organisation legally rely on an algorithm when the algorithm systematically disadvantages an individual or group without adequate legal justification?

In many circumstances, the answer will depend upon who deployed the algorithm, what legal relationship exists, the nature of the affected right, the source of the classification, and whether the differential treatment is legally justified.

2. What Can Constitute an Algorithmic Equity Claim?

An equity claim may arise where an algorithm:

1. rejects loan applications disproportionately affecting a protected group;
2. produces discriminatory recruitment scores;
3. gives different insurance premiums to similarly situated persons;
4. systematically disadvantages persons with disabilities;
5. allocates public benefits unequally;
6. ranks students unfairly;
7. produces discriminatory employee evaluations;
8. uses proxy variables for caste, sex, disability or other protected characteristics;
9. produces unequal access to digital services;
10. makes automated decisions based on inaccurate data;
11. creates discriminatory predictive-policing classifications;
12. denies financial services because of an erroneous risk score.

The legal challenge may concern either intentional discrimination or structural/systemic discrimination, depending on the applicable law.

3. Why Algorithms Create Special Equity Problems

Algorithms may appear neutral because they use mathematical rules.

However, neutrality of code does not necessarily mean neutrality of outcome.

For example:

Historical hiring data → algorithm trained on historical decisions → historical patterns reproduced → applicants from an underrepresented group receive lower scores.

The algorithm may contain no explicit instruction saying:

“Reject women.”

Nevertheless, variables such as:

• employment history;
• career interruptions;
• geographical location;
• educational institution;
• language;
• spending behaviour;
• occupation;
• previous salary;
• postcode;

can function as proxies that correlate with protected characteristics.

Thus, algorithmic discrimination can be:

Direct

The protected characteristic is expressly used.

Indirect

A seemingly neutral criterion disproportionately disadvantages a protected group.

Proxy discrimination

A variable substitutes for a protected characteristic.

Historical discrimination

The model learns discriminatory patterns from historical data.

Structural discrimination

The system's design systematically creates unequal access or outcomes.

4. Constitutional Framework

The Indian Constitution provides the strongest framework for public-sector algorithmic equity claims.

Article 14

Article 14 prohibits:

• arbitrariness;
• irrational classification;
• unequal treatment without adequate justification.

It is particularly important where algorithms are used by:

• government departments;
• public-sector undertakings;
• regulators;
• municipalities;
• public universities;
• welfare authorities;
• police;
• tax authorities.

5. Articles 15 and 16

Article 15

Article 15 prohibits discrimination on specified grounds in constitutionally covered contexts.

Article 16

Article 16 provides equality of opportunity in public employment.

Consequently, algorithmic recruitment or promotion systems used by government departments can potentially be challenged where they create discriminatory outcomes.

6. Article 21

Article 21 is also important because algorithmic systems can affect:

• dignity;
• privacy;
• autonomy;
• livelihood;
• personal liberty;
• informational control.

The Supreme Court's privacy jurisprudence makes algorithmic profiling particularly significant.

7. Algorithmic Equity in Private Organisations

A crucial distinction must be made between public and private actors.

A government algorithm can directly attract constitutional review.

A purely private employer, bank, insurer or technology company does not automatically become subject to every constitutional equality requirement merely because it uses an algorithm.

Private claims may instead arise under:

• employment statutes;
• disability law;
• consumer law;
• contract;
• tort;
• data-protection law;
• sectoral regulations;
• statutory anti-discrimination obligations.

However, where a private entity performs a public function or is otherwise amenable to writ jurisdiction, constitutional principles may become relevant depending on the circumstances.

8. Major Case Laws

The following cases are important authorities for constructing algorithmic-equity arguments. Most are not cases about AI itself; they establish legal principles that can be applied to algorithmic decision-making.

1. E.P. Royappa v State of Tamil Nadu

(1974) 4 SCC 3

Principle

The Supreme Court substantially expanded Article 14 jurisprudence by recognising that arbitrariness is antithetical to equality.

Algorithmic relevance

Suppose an automated government system:

• gives unexplained scores;
• treats similarly situated citizens differently;
• uses irrelevant variables;
• produces inconsistent outcomes.

The affected person may argue that the algorithmic decision is constitutionally arbitrary.

The importance of Royappa is that an equity claim does not necessarily require proof of an explicit discriminatory intention.

9. Maneka Gandhi v Union of India

(1978) 1 SCC 248

Principle

The Supreme Court developed the relationship between Articles 14, 19 and 21 and emphasised that State procedure must be fair, just and reasonable rather than arbitrary.

Algorithmic application

An automated decision affecting a person's:

• licence;
• benefit;
• employment;
• movement;
• financial access; or
• other protected interest

cannot necessarily be justified merely by saying:

“The computer generated the result.”

Where State action significantly affects rights, algorithmic decision-making should be accompanied by appropriate procedural safeguards.

10. Ajay Hasia v Khalid Mujib Sehravardi

(1981) 1 SCC 722

Principle

The case developed the doctrine concerning State action and Article 14, including when bodies that are not formally government departments can nevertheless fall within constitutional scrutiny.

Algorithmic equity relevance

Modern public services are often delivered through:

• government companies;
• statutory corporations;
• public authorities;
• technology contractors.

Therefore, simply outsourcing an algorithmic decision does not necessarily answer the constitutional question.

The relevant inquiry can include the nature and degree of governmental control and the function being performed.

11. Air India v Nergesh Meerza

(1981) 4 SCC 335

Principle

The Supreme Court examined discriminatory employment conditions affecting women and applied constitutional equality principles.

Algorithmic relevance

This case is particularly useful for algorithmic employment discrimination.

Suppose an automated employment system systematically penalises:

• pregnancy-related career breaks;
• maternity leave;
• women returning to work;
• female-dominated occupations.

A claimant could use constitutional sex-equality jurisprudence to challenge discriminatory employment criteria where the relevant constitutional/statutory framework applies.

12. C.B. Muthamma v Union of India

(1979) 4 SCC 260

Principle

The Supreme Court criticised discriminatory service rules affecting women in public employment.

Algorithmic relevance

The case demonstrates that apparently neutral employment administration cannot be structured around gender stereotypes.

If an algorithm learns historical assumptions such as:

“Women are less suitable for leadership positions,”

the use of that historical pattern does not make the resulting discrimination lawful.

The case is therefore valuable for analysing algorithmic gender bias in public employment.

13. Jeeja Ghosh v Union of India

(2016) 7 SCC 761

Principle

The Supreme Court strongly emphasised dignity, equality and the rights of persons with disabilities.

Algorithmic relevance

AI systems can disadvantage persons with disabilities through:

• inaccessible interfaces;
• speech-dependent systems;
• facial-recognition systems;
• automated recruitment tests;
• productivity monitoring;
• standardised assessments;
• failure to provide reasonable accommodation.

The case supports a rights-based approach in which formal equality is not always enough.

14. Vikash Kumar v Union Public Service Commission

(2021) 5 SCC 370

Principle

The Supreme Court recognised the importance of reasonable accommodation under disability-rights law.

Algorithmic relevance

This is one of the strongest authorities for algorithmic equity involving disability.

An automated examination or recruitment system may produce an apparently identical requirement for everyone but disproportionately disadvantage a person with a disability.

The appropriate legal question may therefore be:

Did the system provide reasonable accommodation necessary to achieve substantive equality?

An algorithm designed around a “one-size-fits-all” model may fail this requirement.

15. Rajive Raturi v Union of India

(2024) 6 SCC 418

Importance

The Supreme Court addressed accessibility and the rights of persons with disabilities.

Algorithmic relevance

Algorithmic equity cannot be confined to whether the mathematical model treats users identically.

A system may be formally identical but practically inaccessible.

For example:

An online government benefit system requiring visual interaction without an accessible alternative may technically apply the same rule to everyone while substantially excluding persons with visual disabilities.

The principle of substantive accessibility therefore becomes highly relevant to algorithmic systems.

16. National Federation of the Blind v UPSC

(2013) 10 SCC 772

Principle

The Supreme Court protected equality of opportunity for persons with visual disabilities in public employment.

Algorithmic relevance

Automated recruitment and assessment systems must not create technological barriers that effectively exclude persons with disabilities.

The case supports the proposition that digital equality must include meaningful access, not merely formal permission to participate.

17. K.S. Puttaswamy v Union of India

(2017) 10 SCC 1

Principle

The Supreme Court recognised privacy as a fundamental right.

Privacy includes dimensions of:

• autonomy;
• dignity;
• informational privacy;
• control over personal information.

Algorithmic equity relevance

Algorithmic profiling may use:

• financial information;
• location;
• browsing behaviour;
• biometric information;
• employment history;
• health information;
• social data.

Where profiling produces discriminatory consequences, the privacy and equality dimensions can overlap.

For example:

Excessive collection → sensitive profiling → discriminatory classification → denial of service.

Thus, an algorithmic equity claim may also be a privacy claim.

18. K.S. Puttaswamy (Aadhaar) v Union of India

(2019) 1 SCC 1

Principle

The Supreme Court considered:

• proportionality;
• privacy;
• authentication;
• exclusion;
• data architecture;
• constitutional safeguards.

Algorithmic relevance

The case is particularly significant for automated identity and eligibility systems.

A technological system can produce exclusion because of:

• authentication failures;
• inaccurate data;
• database errors;
• biometric mismatch;
• incorrect classification.

The broader lesson is:

Technological efficiency cannot automatically override constitutional rights or produce unjust exclusion.

19. Anuj Garg v Hotel Association of India

(2008) 3 SCC 1

Principle

The Supreme Court examined gender stereotypes and held that laws based on paternalistic assumptions concerning women require constitutional scrutiny.

Algorithmic relevance

This is particularly important for machine-learning systems trained on historical social patterns.

If an algorithm effectively assumes:

“Women should not perform this occupation,”

or:

“Women are less suited to certain roles,”

historical social practices cannot by themselves supply constitutional justification.

The case is therefore useful in analysing stereotype-driven algorithmic discrimination.

20. Navtej Singh Johar v Union of India

(2018) 10 SCC 1

Principle

The Supreme Court strongly emphasised dignity, equality, autonomy and protection against discrimination.

Algorithmic relevance

Although not an algorithm case, it supports broader constitutional principles against identity-based exclusion.

Algorithmic systems that classify individuals based on sensitive personal characteristics therefore require especially careful scrutiny.

21. National Legal Services Authority v Union of India

(2014) 5 SCC 438

Principle

The Supreme Court recognised constitutional protection for transgender persons and emphasised equality, dignity and identity.

Algorithmic relevance

Automated systems can produce exclusion when their categories do not accommodate diverse identities.

Examples include:

• binary-only forms;
• automated identity verification;
• employment databases;
• benefit systems;
• educational portals.

A system that technically processes data but excludes a legally protected identity may create a substantive-equality problem.

22. Shreya Singhal v Union of India

(2015) 5 SCC 1

Principle

The Supreme Court struck down Section 66A of the Information Technology Act for violating constitutional freedoms and also considered intermediary-related questions.

Algorithmic relevance

The case is relevant to the broader proposition that technology does not receive immunity from constitutional scrutiny.

A government cannot transform an unconstitutional restriction into a lawful one simply by implementing it through a computer system.

23. Anuradha Bhasin v Union of India

(2020) 3 SCC 637

Principle

Restrictions affecting fundamental rights must satisfy requirements including legality, necessity and proportionality.

Algorithmic relevance

Where an algorithm is used to restrict:

• access;
• communications;
• financial services;
• digital platforms;
• public benefits;

the proportionality of the resulting restriction may become relevant.

24. A.K. Kraipak v Union of India

(1969) 2 SCC 262

Principle

The Supreme Court significantly developed the doctrine of natural justice in administrative decision-making.

Algorithmic relevance

An algorithmic decision should not automatically eliminate:

• impartiality;
• procedural fairness;
• opportunity to respond;
• meaningful review.

Where an automated decision has serious consequences, human oversight may become particularly important.

25. State of Orissa v Dr. Binapani Dei

AIR 1967 SC 1269

Principle

Administrative decisions having civil consequences must observe appropriate procedural fairness.

Algorithmic relevance

If an automated system produces a decision affecting a person's legal or economic position, the affected person may be entitled to procedural safeguards appropriate to the statutory context.

26. Algorithmic Equity in Employment

Employment is one of the most significant areas.

AI may determine:

• recruitment ranking;
• interview selection;
• employee performance;
• promotion;
• compensation;
• attendance;
• productivity;
• termination risk.

Example

An employer trains an AI recruitment model on ten years of historical employee data.

Historically, management selected men disproportionately for senior positions.

The algorithm learns that pattern.

Consequently:

male candidate → higher predicted leadership score
female candidate → lower predicted leadership score.

The employer may argue:

“The algorithm was gender-neutral.”

But the legal question is not simply whether the word “male” appeared in the code.

The inquiry can involve:

• the source data;
• proxy variables;
• discriminatory impact;
• applicable employment law;
• constitutional obligations if the employer is a State/public authority;
• disability/gender statutes;
• contractual duties.

27. Algorithmic Equity in Banking and Credit

Banks increasingly use automated models for:

• credit scoring;
• fraud detection;
• loan approval;
• customer risk;
• anti-money-laundering screening.

Potential problems include:

• discriminatory credit scoring;
• geographic proxies;
• incorrect financial data;
• unexplained rejection;
• automated account closure;
• excessive risk classification.

A claimant may need to demonstrate:

Algorithmic classification → unequal treatment → legally protected interest/duty → unjustified differentiation → injury.

28. Algorithmic Equity in Insurance

Insurance algorithms may assess:

• health risk;
• driving behaviour;
• location;
• financial history;
• claims history;
• predicted probability of loss.

Potential equity issues include:

• discriminatory pricing;
• exclusion;
• proxy discrimination;
• use of sensitive information;
• inaccurate risk scores.

The legality depends substantially on applicable insurance regulations and the specific contractual/statutory framework.

29. Algorithmic Equity in Public Welfare

Government algorithms may determine eligibility for:

• pensions;
• food assistance;
• scholarships;
• housing;
• healthcare;
• employment schemes;
• subsidies.

The stakes are especially high because an erroneous classification may affect basic rights or entitlements.

A claimant may challenge:

• erroneous databases;
• opaque eligibility criteria;
• lack of notice;
• inability to correct data;
• automated rejection;
• lack of human review.

The Puttaswamy-Aadhaar jurisprudence is particularly relevant to technological exclusion.

30. Algorithmic Equity and Disability

This is a particularly important category.

Formal equality

Everyone uses exactly the same system.

Substantive equality

The system is modified where necessary so that persons with disabilities can participate meaningfully.

Indian disability jurisprudence strongly supports the latter approach.

Therefore:

“The algorithm treats everyone identically”

does not necessarily defeat an equality claim.

An inaccessible algorithm can itself produce discriminatory exclusion.

31. Algorithmic Equity and Data Protection

The Digital Personal Data Protection Act, 2023, together with the applicable rules and commencement framework, is relevant to algorithmic systems processing digital personal data.

Potential concerns include:

• unlawful processing;
• excessive collection;
• inaccurate personal data;
• security failures;
• unauthorised disclosure;
• incompatible processing;
• failure to comply with applicable data-subject rights.

Data protection and equality are different legal concepts, but they can overlap.

For example:

inaccurate personal data → inaccurate algorithmic profile → discriminatory decision → financial/employment harm.

32. Direct vs Indirect Algorithmic Discrimination

Direct discrimination

An algorithm explicitly uses sex, caste or another protected characteristic.

Indirect discrimination

The algorithm uses a neutral criterion that disproportionately disadvantages a protected group.

Proxy discrimination

A variable indirectly represents a protected characteristic.

For example:

postcode → strongly correlated with caste/community → lower credit score.

Historical-data discrimination

The algorithm learns existing social inequalities.

This is one of the greatest risks of machine-learning systems.

33. Evidentiary Problems

Algorithmic equity cases often face a difficult evidentiary question:

How does the claimant prove that an algorithm discriminated against them?

Relevant evidence may include:

• decision outputs;
• datasets;
• model documentation;
• feature lists;
• model versions;
• validation reports;
• error rates;
• group-impact statistics;
• audit records;
• human-review records;
• comparator cases;
• internal communications;
• vendor agreements;
• testing results.

Courts may also need to distinguish between:

correlation and legally relevant discrimination.

A statistical disparity can be important evidence, but it does not automatically establish liability in every legal context.

34. The Black-Box Problem

A major issue is algorithmic opacity.

An affected person may know:

“My application was rejected.”

But not know:

• why;
• what data was used;
• which variables mattered;
• whether the information was accurate;
• whether similarly situated people were treated differently.

This creates a procedural problem.

The legal system traditionally expects decisions affecting rights to be sufficiently explainable to permit meaningful challenge.

The principles in Mohinder Singh Gill and S.N. Mukherjee become particularly useful in this context.

35. Human Oversight

Human oversight should not be purely symbolic.

A system in which an employee merely clicks:

“Approve algorithmic result”

may not constitute meaningful review.

Effective review should permit the human decision-maker to:

• inspect relevant information;
• identify obvious errors;
• consider exceptional circumstances;
• override an incorrect result;
• document reasons.

This is especially important where decisions affect:

• employment;
• liberty;
• public benefits;
• disability rights;
• credit;
• essential services.

36. Remedies

Depending upon the legal framework, remedies may include:

Constitutional remedies

Under Articles 32 and 226:

• writ of mandamus;
• certiorari;
• prohibition;
• declaration;
• directions for reconsideration.

Administrative remedies

• human review;
• correction of records;
• reconsideration;
• re-evaluation.

Statutory remedies

Where provided by legislation:

• complaints;
• appeals;
• regulatory proceedings;
• compensation.

Civil remedies

Depending upon the cause of action:

• damages;
• injunction;
• declaration;
• specific relief.

37. Practical Legal Test

A useful framework for an Indian algorithmic-equity claim is:

Step 1 — Identify the decision

What did the algorithm actually decide?

Step 2 — Identify the decision-maker

Was it:

• government;
• public authority;
• employer;
• bank;
• insurer;
• platform;
• private company?

Step 3 — Identify the protected interest

Is the affected interest related to:

• equality;
• employment;
• disability;
• privacy;
• property;
• livelihood;
• consumer rights;
• contractual rights?

Step 4 — Identify the classification

What variable caused differential treatment?

Step 5 — Identify the comparator

Who was treated differently?

Step 6 — Establish disparity

Is there actual unequal treatment or exclusion?

Step 7 — Examine justification

Is the differentiation legally authorised and objectively justified?

Step 8 — Examine proportionality

Is the measure necessary and appropriately tailored?

Step 9 — Examine procedure

Was notice, explanation, hearing or review required?

Step 10 — Examine causation

Did the algorithmic decision actually cause the legal injury?

38. Important Case-Law Matrix

39. Core Legal Formula

A useful formulation for litigation is:

Algorithmic Classification + Differential Treatment + Protected Legal Interest + Lack of Adequate Justification/Procedural Safeguard + Causally Connected Injury = Potential Algorithmic Equity Claim

However, the precise elements depend on the legal cause of action.

40. Key Distinction: Unfairness vs Illegality

Not every unfair algorithmic outcome automatically creates a successful legal claim.

A claimant generally needs to connect the outcome to a recognised legal obligation.

For example:

“The AI gave me a low score.”

by itself may not establish liability.

But:

“The AI gave me a lower score because of a prohibited discriminatory criterion, and that score caused denial of a legally protected opportunity”

is substantially stronger.

Similarly:

“The model has a 5% higher error rate for Group A”

is evidence of possible inequity, but the legal consequence depends upon the applicable statute, constitutional provision, duty and facts.

41. Conclusion

Algorithmic Equity Claims in India are an emerging field rather than a standalone cause of action. Their legal foundation comes primarily from the constitutional principles of equality, non-arbitrariness, dignity, privacy, substantive equality and reasonable accommodation, supplemented by employment, disability, consumer, contract, data-protection and sector-specific law.

The most important authorities include:

1. E.P. Royappa v State of Tamil Nadu
2. Maneka Gandhi v Union of India
3. A.K. Kraipak v Union of India
4. C.B. Muthamma v Union of India
5. Air India v Nergesh Meerza
6. Jeeja Ghosh v Union of India
7. Vikash Kumar v UPSC
8. Rajive Raturi v Union of India
9. NALSA v Union of India
10. Anuj Garg v Hotel Association of India
11. Puttaswamy v Union of India
12. Puttaswamy (Aadhaar) v Union of India
13. S.N. Mukherjee v Union of India
14. Anuradha Bhasin v Union of India

The central principle is:

An algorithm's mathematical neutrality does not automatically establish legal equality. If an automated system produces arbitrary, discriminatory, inaccessible or disproportionate outcomes, the legality of the underlying classification, data, decision-making process and safeguards can be judicially examined.

In particular, public authorities cannot avoid Article 14 or Article 21 simply by transferring a decision from a human officer to an algorithm, and disability-related systems may additionally have to satisfy the substantive-equality and reasonable-accommodation requirements recognised by the Supreme Court.