AI-DRIVEN HIRING TOOLS: DISCRIMINATION RISK IN PAKISTAN

1. Introduction

Artificial Intelligence (AI) is increasingly being used in recruitment for CV screening, candidate ranking, automated aptitude testing, video-interview analysis, personality assessment and shortlisting. Although these tools can improve speed and reduce administrative costs, they may also reproduce or amplify discrimination contained in historical recruitment data.

In Pakistan, there is presently no comprehensive statute specifically regulating discriminatory AI recruitment systems. Therefore, the legality of AI-driven hiring must largely be examined through constitutional equality principles, employment laws, applicable service rules, privacy principles and existing judicial decisions concerning discriminatory employment practices.

The most important constitutional provisions are Articles 25 and 27 of the Constitution of Pakistan, 1973. Article 25 guarantees equality before law and equal protection of law and prohibits discrimination on the basis of sex, while Article 27 specifically protects qualified citizens against discrimination in appointments to the service of Pakistan on specified grounds.

2. Meaning of AI-Driven Hiring Discrimination

AI-driven hiring discrimination occurs when an automated recruitment system produces systematically unequal or disadvantageous outcomes for candidates because of characteristics such as:

sex or gender;

race or ethnicity;

religion;

caste;

place of birth or residence;

disability;

socioeconomic background;

age;

language;

educational background; or

other characteristics that operate as proxies for protected characteristics.

For example, an AI recruitment system may learn from historical company data in which predominantly male candidates were hired for senior technical positions. If the system is trained on that historical data, it may learn to rank male candidates more highly even though gender is not expressly included as an input.

Thus, discrimination may occur without an employer intentionally instructing the AI system to discriminate.

3. Constitutional Framework in Pakistan

Article 25 — Equality

Article 25(1) provides that all citizens are equal before the law and entitled to equal protection of law. Article 25(2) provides that there shall be no discrimination on the basis of sex, while Article 25(3) permits special provisions for the protection of women and children.

Consequently, an AI recruitment system that systematically disadvantages qualified candidates because of gender could raise serious constitutional equality concerns, particularly where the recruitment is undertaken by the State or a public authority.

Article 27 — Non-discrimination in Public-Service Appointment

Article 27 is particularly relevant to recruitment. It provides that a citizen otherwise qualified for appointment in the service of Pakistan shall not be discriminated against in respect of appointment on grounds including race, religion, caste, sex, residence or place of birth, subject to constitutional exceptions concerning representation and specified posts.

Therefore, an algorithm used by a public-sector institution cannot automatically escape constitutional scrutiny merely because the discriminatory decision was technically produced by software.

4. Major Risks of AI Hiring Systems

A. Historical Data Bias

AI systems learn patterns from previous recruitment decisions. If previous hiring decisions contained discrimination, the algorithm may reproduce those patterns.

For example, if a company historically hired men for engineering positions, the algorithm may incorrectly treat characteristics associated with previous male employees as indicators of suitability.

This creates the possibility of algorithmic discrimination without deliberate human discrimination.

B. Gender Discrimination

Gender discrimination is especially important under Pakistani constitutional law because Article 25 expressly addresses sex discrimination and Article 27 prohibits discrimination in public-service appointments on the ground of sex, subject to its constitutional exceptions.

An AI system could discriminate indirectly through:

employment gaps;

maternity-related career interruptions;

gender-associated names;

participation in women-focused organizations;

historical salary information; or

patterns in previous male-dominated hiring.

The system might not explicitly receive "female" as an input but could still use variables that function as proxies for gender.

C. Religious Discrimination

Pakistan's constitutional framework expressly prohibits certain forms of discrimination in public-service appointment based on religion. An AI system trained on biased historical recruitment data could potentially disadvantage candidates associated with a particular religious group.

Such automated differentiation would require careful legal scrutiny because algorithmic neutrality cannot be established merely by removing an explicit religion field.

D. Residence and Regional Bias

Residence and location can become problematic when an algorithm uses:

residential address;

postcode;

city;

province;

university location; or

previous employment location

as predictive variables.

Article 27 expressly refers to residence and place of birth in the context of discrimination in public-service appointments, while the Constitution also permits certain qualifications and representation measures.

Therefore, employers must distinguish between lawful recruitment classifications and arbitrary exclusion.

E. Proxy Discrimination

One of the most important risks is proxy discrimination.

A system may not directly use gender, religion or ethnicity but may use variables strongly correlated with them.

For example:

Gender → college attended → geographical location → employment history → AI score

If the final AI score consistently disadvantages women because the system relies on variables correlated with gender, simply removing the "gender" field may not eliminate discrimination.

5. Lack of Transparency

Many AI hiring systems operate through complex models whose decision-making processes may be difficult for recruiters and applicants to understand.

A rejected applicant may therefore receive only a statement that the candidate did not meet the required "AI score."

This creates several legal problems:

The candidate may not know why he or she was rejected.

The employer may not know whether the algorithm contains discriminatory patterns.

It becomes difficult to challenge an incorrect decision.

Human decision-makers may place excessive reliance on an automated recommendation.

Evidence concerning discriminatory treatment may become difficult to obtain.

For public-sector recruitment, these concerns are particularly significant because constitutional equality and merit requirements apply to governmental appointments.

6. Employer Liability for AI Decisions

An important legal principle is that an employer should not automatically avoid responsibility merely because an AI system made the initial recommendation.

The legal question should focus on:

Who selected the system, supplied the data, established the recruitment criteria, relied upon the output and ultimately made the appointment decision?

If an employer knowingly relies on an algorithm that produces discriminatory results, the use of technology may not provide a defence against applicable equality requirements.

Accordingly, employers should conduct:

pre-deployment bias testing;

regular validation;

impact assessments;

human review;

documentation of recruitment criteria;

audit trails; and

procedures for candidate complaints.

7. Relevant Pakistani Case Laws

Case Law 1: General Post Office, Islamabad v. Muhammad Jalal

PLD 2024 Supreme Court 1276

The Supreme Court considered government recruitment practices under Articles 18, 25 and 27. The Court held that appointments made without open advertisement, competition and merit could unlawfully exclude other citizens and found such practices inconsistent with constitutional equality and equal employment opportunity.

Relevance to AI Hiring

This case is highly relevant by analogy. If an AI recruitment system effectively excludes qualified candidates through an undisclosed or discriminatory algorithm, the employer may face questions concerning:

equality;

merit;

transparency;

fair competition; and

arbitrary exclusion.

The case does not concern AI specifically, but its constitutional principles can be applied when evaluating automated recruitment practices.

Case Law 2: Secretary Finance, Government of Khyber Pakhtunkhwa v. Syed Jehangir Shah

2024 SCMR 538

The Supreme Court considered Article 25 and discriminatory treatment in public employment. The Court emphasized that classifications affecting similarly situated employees must have a reasonable basis rather than being arbitrary.

Relevance to AI Hiring

An AI system that places candidates into different categories must have objectively defensible recruitment criteria.

For example, if two candidates have substantially similar qualifications but the algorithm consistently gives lower scores to one group without a legitimate employment-related reason, the classification may require constitutional scrutiny.

Case Law 3: Muhammad Yaseen v. Secretary, Ministry of Interior

2023 SCMR 1691

The Supreme Court addressed discriminatory treatment in government service under Article 25. The Court recognized that Article 25 prohibits discriminatory treatment and required similarly situated employees to be treated consistently.

Relevance to AI Hiring

The principle supports the proposition that recruitment systems should not arbitrarily treat similarly situated applicants differently.

An AI model that generates materially different outcomes for comparable candidates without a legitimate and explainable reason may therefore create an equality concern.

Case Law 4: Chief Secretary, Government of Balochistan v. Adeel-ur-Rehman

2024 SCMR 142 / 2024 SCLR 26

The Supreme Court explained that Article 25 does not create an entitlement to "negative equality." It also considered the legality of classification in public employment and emphasized the importance of lawful criteria for recruitment and appointment.

Relevance to AI Hiring

The case demonstrates that equality does not mean that every candidate must receive identical treatment. A recruitment algorithm can legitimately distinguish between candidates when the distinction is supported by lawful and relevant criteria.

The problem arises where the distinction is arbitrary, irrelevant or discriminatory.

Case Law 5: Nadia Naz v. President of Pakistan

PLD 2023 SC 588

The Supreme Court interpreted workplace harassment broadly to include gender-based discrimination and emphasized equal opportunity and equal treatment in employment. The Court recognized that gender-based discriminatory conduct can interfere with workplace rights and dignity.

Relevance to AI Hiring

Although the case concerns workplace harassment rather than recruitment algorithms, it demonstrates the Supreme Court's recognition that gender-based discrimination can exist beyond overtly sexual conduct.

An AI recruitment mechanism that systematically disadvantages candidates on gender-related grounds could therefore raise broader equality and dignity concerns.

Case Law 6: Civil Aviation Authority v. Union of Civil Aviation Employees

PLD 1997 SC 781

This decision is among the cases cited by Pakistani courts in relation to constitutional and employment-law questions concerning equality and public employment. A Sindh High Court judgment discussing Article 27 specifically referred to the case along with other Supreme Court authorities concerning constitutional review and equality.

Relevance to AI Hiring

The case is useful for understanding that employment decisions of public authorities remain subject to constitutional standards. Delegating part of the recruitment process to technology does not, by itself, remove the underlying legal obligations of the public employer.

Case Law 7: Elahi Cotton Mills Ltd. v. Federation of Pakistan

PLD 1997 SC 582

Elahi Cotton Mills is a leading constitutional authority concerning equality and reasonable classification. Pakistani courts have repeatedly referred to it when considering whether a legal classification has a rational and constitutionally permissible basis.

Relevance to AI Hiring

The principles concerning reasonable classification can be applied when assessing algorithmic recruitment criteria.

An AI system should therefore distinguish candidates using legitimate, job-related factors rather than arbitrary characteristics or discriminatory proxies.

8. Private-Sector and Public-Sector Distinction

The strongest direct constitutional application of Article 27 concerns appointments in the service of Pakistan.

For private employers, the legal analysis is broader and may involve applicable labour legislation, contractual obligations, anti-harassment legislation, constitutional principles where applicable, and general protections against discriminatory treatment.

Therefore, an AI recruitment system used by:

a federal department;

provincial department;

government-owned entity;

public authority; or

private company

may raise somewhat different legal questions.

The public-sector employer faces particularly direct Article 27 considerations.

9. Data-Protection Dimension

AI recruitment systems generally require substantial amounts of personal information, including CVs, employment history, educational qualifications, interview recordings and potentially biometric or behavioural information.

Pakistan's Ministry of IT and Telecommunication currently lists the Personal Data Protection Bill as draft legislation rather than an approved statute on its legislation portal. The portal identifies the 2020 and May 2023 versions as drafts.

Therefore, as of September 2026, employers should not describe a proposed data-protection framework as though it were automatically equivalent to an enacted comprehensive personal-data protection statute.

This is important for AI recruitment because employers should nevertheless adopt privacy safeguards such as:

collecting only necessary information;

informing applicants about data use;

securing recruitment databases;

restricting access;

retaining data only as necessary;

preventing unauthorized secondary use; and

documenting automated decision-making processes.

10. Human Oversight

A major safeguard against AI discrimination is human-in-the-loop decision-making.

AI should ideally assist rather than independently determine employment eligibility.

A responsible recruitment structure could be:

Candidate Application → AI Screening → Bias/Quality Check → Human Review → Interview → Final Decision

This allows recruiters to investigate unusual outcomes and prevents an automated score from becoming an unquestioned employment decision.

11. Algorithmic Auditing

Employers using AI hiring systems should periodically examine:

selection rates by gender;

selection rates across relevant demographic groups;

false rejection rates;

interview-selection rates;

differences in AI scores;

data-quality problems;

proxy variables;

model drift; and

complaints from applicants.

An algorithm that was initially neutral may become discriminatory when recruitment data, job requirements or candidate populations change.

12. Remedies for Discriminatory AI Recruitment

A candidate affected by discriminatory automated recruitment may potentially seek relief through the forum and legal mechanism applicable to the employer and employment relationship.

Possible forms of relief may include:

reconsideration of the recruitment decision;

disclosure or explanation of relevant recruitment criteria where legally available;

constitutional judicial review in appropriate public-sector cases;

correction of discriminatory recruitment practices;

fresh recruitment or reconsideration;

appropriate employment-law remedies; and

institutional directions requiring fair and lawful recruitment procedures.

The exact remedy depends on the employer's legal status, applicable service rules and facts of the individual case.

13. Compliance Framework for Pakistani Employers

Employers adopting AI hiring systems should consider the following safeguards:

1. Bias Testing

Test the model before deployment to identify discriminatory outcomes.

2. Representative Training Data

Use recruitment data that does not systematically reproduce historical discrimination.

3. Job-Related Criteria

AI variables should be demonstrably connected to genuine requirements of the position.

4. Human Review

Final recruitment decisions should receive meaningful human oversight.

5. Explainability

Recruiters should be able to understand the principal factors affecting candidate ranking.

6. Audit Trails

Maintain records showing how recruitment decisions were generated.

7. Candidate Complaint Mechanism

Applicants should have an avenue for challenging apparent errors or discriminatory outcomes.

8. Periodic Audits

The system should be tested periodically rather than only once before deployment.

9. Privacy Protection

Candidate information should be collected and processed responsibly.

10. Public-Sector Constitutional Compliance

Government employers should ensure that automated recruitment remains consistent with Articles 18, 25 and 27 and applicable service rules.

14. Hypothetical Example

Suppose a Pakistani company introduces an AI recruitment system for software engineers.

The system examines:

CV;

university;

previous employment;

employment gaps;

interview answers;

location;

language;

online assessment score.

After six months, an audit discovers that qualified female candidates are shortlisted at a substantially lower rate than similarly qualified male candidates.

The employer cannot necessarily defend the system simply by stating:

"The computer made the decision."

The appropriate legal inquiry would include:

What data trained the model?

Were historical recruitment decisions biased?

Were gender proxies used?

Were the variables genuinely job-related?

Was human review available?

Were candidates given a meaningful opportunity to challenge errors?

Was the recruitment process consistent with applicable equality and employment requirements?

15. Conclusion

AI-driven hiring can improve efficiency, but automation does not eliminate legal responsibility for discriminatory recruitment. In Pakistan, the principal constitutional safeguards are found in Articles 25 and 27, supported by judicial principles concerning equality, reasonable classification, merit and non-discriminatory public employment.

There does not appear to be a reported Pakistani Supreme Court decision specifically deciding the legality of an AI hiring algorithm as such. Consequently, the existing case law must presently be applied by analogy to automated recruitment.

The central legal principle is that technology cannot be used as a mechanism for avoiding equality obligations. A Pakistani employer using AI for recruitment should therefore ensure that its system is transparent, job-related, periodically audited, subject to human oversight and capable of detecting and correcting discriminatory outcomes.

In short, AI may automate recruitment, but it should not automate discrimination.