Explainability Reports for HR AI — Detailed Legal Explanation

1. Meaning of Explainability Reports for HR AI

An explainability report for HR AI is a document that explains, in understandable and legally defensible terms, how and why an artificial-intelligence system reached an employment-related decision or recommendation.

HR AI may be used for:

recruitment and résumé screening;

candidate ranking;

interview scoring;

promotion recommendations;

performance assessment;

compensation recommendations;

employee attrition prediction;

disciplinary-risk assessment;

workforce allocation;

termination or redundancy recommendations; and

internal mobility and succession planning.

An explainability report is particularly important where the AI output has a significant effect on an individual's employment opportunity, status, pay, promotion, or continuation of employment.

The central legal concern is not simply whether the AI system is technically accurate. It is whether the employer can demonstrate that the decision was:

lawful;

non-discriminatory;

based on relevant criteria;

procedurally fair;

supported by reliable evidence;

capable of meaningful human review; and

properly documented.

2. Why Explainability Matters in HR AI

Traditional HR decisions can generally be explained by identifying the decision-maker and the reasons relied upon.

For example:

"The employee was denied promotion because she did not satisfy the required performance criteria."

With AI, an employer may instead say:

"The algorithm assigned the employee a promotion score of 61."

That answer is legally inadequate in many circumstances.

The important questions become:

What factors produced the score?

Which factors were given greater weight?

Was the data accurate?

Was the training data representative?

Did the algorithm indirectly use protected characteristics?

Was historical discrimination embedded in the training data?

Was the employee given an opportunity to challenge inaccurate information?

Did a human independently review the recommendation?

Was the algorithm validated?

Was the same model applied consistently to similarly situated employees?

An explainability report helps answer these questions.

3. What Should an HR AI Explainability Report Contain?

A robust report should normally contain the following components.

A. Purpose of the AI system

The employer should identify:

what the system is designed to do;

what employment decision it supports;

whether it makes a recommendation or an actual decision;

who is authorised to use it; and

what risks are associated with its use.

For example:

"The system ranks applicants for initial recruitment screening. It does not independently determine who will be hired."

This distinction is extremely important.

B. Data used by the system

The report should identify the categories of data used, such as:

education;

qualifications;

work experience;

performance history;

attendance;

skills;

job-related assessments;

interview results; and

other legitimately relevant employment information.

It should also identify data that should not influence the decision, such as protected characteristics where applicable.

C. Model methodology

The report should explain, at an appropriate level:

what type of AI/model was used;

how the model was trained;

what variables were considered;

how variables were weighted;

how the model was validated;

how often it is tested; and

known limitations.

The employer does not necessarily have to disclose proprietary source code.

However, trade secrecy should not become a blanket excuse for withholding legally necessary explanations.

4. Global Explanation vs Individual Explanation

One of the most important distinctions is between system-level explainability and individual-decision explainability.

System-level explanation

This explains:

"How does our recruitment AI generally work?"

For example:

résumé information is analysed;

relevant qualifications are identified;

experience is compared against job requirements;

candidates receive a ranking;

recruiters conduct human review.

Individual-level explanation

This explains:

"Why did this particular candidate receive this particular result?"

For example:

"The applicant's ranking was primarily affected by insufficient experience in the two mandatory technical skills specified in the job description."

The second explanation is generally much more useful when an individual challenges an employment decision.

5. Counterfactual Explanation

A sophisticated explainability report may also provide a counterfactual explanation.

It answers:

"What would have needed to be different for the outcome to change?"

For example:

"The candidate would have met the screening threshold if the mandatory certification requirement had been satisfied."

This is useful because it allows the affected person and the decision-maker to determine whether the AI relied on a legitimate job-related factor.

However, counterfactual explanations must themselves be carefully designed. An employer should not imply that an individual could have obtained a different result merely by changing an immutable or protected characteristic.

6. Human Oversight

An explainability report should identify:

who reviewed the AI output;

whether the reviewer could override the AI;

whether the reviewer actually exercised independent judgment;

what evidence the reviewer considered;

whether reasons were recorded; and

whether an appeal mechanism existed.

A nominal human reviewer is not necessarily meaningful human oversight.

If the HR manager simply accepts every AI recommendation, the organisation may effectively be operating an automated decision-making system despite formally describing it as human-assisted AI.

7. Bias and Discrimination Testing

This is one of the most important elements of an HR AI explainability report.

The employer should examine whether the AI produces materially different outcomes across relevant groups.

Testing can examine:

selection rates;

rejection rates;

promotion rates;

compensation recommendations;

performance scores;

disciplinary-risk scores; and

termination recommendations.

The organisation should also investigate proxy discrimination.

For example, an algorithm may not explicitly use gender but might rely heavily on variables correlated with gender.

Similarly, location, employment gaps, educational institutions, language patterns or other apparently neutral variables may sometimes function as proxies.

8. Accuracy and Data Quality

Explainability is closely connected with data quality.

An AI system may produce a perfectly consistent result from inaccurate information.

For example:

Employee A's performance database incorrectly records three missed deadlines.

If the AI uses that database and recommends Employee A for termination, explaining the algorithm does not solve the underlying problem.

Therefore, the report should address:

source of data;

accuracy;

completeness;

updating procedures;

correction mechanisms;

retention periods; and

procedures for disputed information.

9. Explainability and Natural Justice

In Indian employment law, the principles of natural justice are particularly important where an employment decision has adverse consequences.

The basic principles include:

Audi alteram partem

A person affected by an adverse decision should generally have a meaningful opportunity to respond where the law requires procedural fairness.

Reasoned decision-making

A decision affecting rights or legitimate interests should ordinarily disclose adequate reasons.

AI therefore creates a potential problem if the employer cannot explain the basis of the decision.

A statement such as:

"The AI system determined that you were unsuitable"

is unlikely to be a satisfactory substitute for meaningful reasons where reasons are legally required.

10. Important Indian Constitutional Framework

Article 14

Article 14 protects against arbitrary state action and requires equality before law and equal protection of laws.

Where a public employer uses AI, arbitrary or irrational algorithmic decision-making can potentially attract Article 14 scrutiny.

Article 16

Article 16 guarantees equality of opportunity in matters of public employment.

Therefore, AI-based recruitment or promotion systems used by government bodies must be particularly carefully designed.

Article 21

Article 21 protects life and personal liberty and has been interpreted broadly to include important aspects of dignity, privacy and procedural fairness.

Where HR AI processes extensive personal information, privacy considerations become relevant.

11. Important Case Laws

The following cases are particularly useful for understanding the legal principles that can apply to explainability, automated decision-making, fairness, privacy and reasoned employment decisions.

Case 1: Maneka Gandhi v. Union of India, (1978) 1 SCC 248

Principle

The Supreme Court significantly expanded the understanding of Article 21 and emphasised that procedure affecting fundamental rights must satisfy standards of fairness and non-arbitrariness.

The Court rejected the idea that merely following a formally prescribed procedure automatically makes State action constitutionally valid.

Relevance to HR AI

If a government employer uses an AI system to make or substantially influence an adverse employment decision, the organisation may need to demonstrate that the process is fair and non-arbitrary.

An explainability report can assist by showing:

what information was considered;

what criteria were applied;

whether the process was consistent;

whether the person had an opportunity to contest errors; and

whether meaningful human review occurred.

Key lesson

Automated procedure cannot be treated as automatically fair merely because it is technologically sophisticated.

12. E.P. Royappa v. State of Tamil Nadu, (1974) 4 SCC 3

Principle

The Supreme Court connected equality under Article 14 with the prohibition of arbitrariness.

The judgment is frequently associated with the principle that arbitrariness is antithetical to equality.

HR AI relevance

Suppose an AI recruitment system consistently rejects candidates based upon an unexplained scoring mechanism.

The employer may be required to demonstrate that:

the criteria are rational;

the criteria have a reasonable connection with the job;

similarly situated candidates are treated consistently; and

the result is not arbitrary.

An explainability report can become evidence of the rational basis of the system.

Key lesson

AI-generated outcomes should have a rational and job-related basis rather than being accepted merely because "the algorithm said so."

13. Ajay Hasia v. Khalid Mujib Sehravardi, (1981) 1 SCC 722

Principle

The Supreme Court emphasised that Article 14 strikes at arbitrariness and that State action must meet constitutional standards of equality.

HR AI relevance

Government-controlled organisations and public employment processes cannot escape constitutional scrutiny simply by delegating decision-making to a computerised or algorithmic system.

If a public authority says:

"The vendor's AI made the selection decision, not us,"

that does not necessarily eliminate the authority's legal responsibility.

Key lesson

Delegating decision-making to an AI vendor does not necessarily eliminate the employer's responsibility for the resulting decision.

14. Shrilekha Vidyarthi v. State of U.P., (1991) 1 SCC 212

Principle

The Supreme Court held that State action, including action involving contractual relationships, can be subject to Article 14 requirements and must not be arbitrary.

HR AI relevance

This is significant because employment relationships increasingly involve automated HR systems.

An employer cannot necessarily avoid legal scrutiny simply by characterising an AI-driven employment decision as an internal contractual or managerial decision, particularly where public law principles apply.

Key lesson

The form of the decision does not automatically determine whether constitutional fairness requirements apply.

15. Justice K.S. Puttaswamy (Retd.) v. Union of India, (2017) 10 SCC 1

Principle

The Supreme Court recognised privacy as a constitutionally protected fundamental right under Article 21.

The judgment recognised important dimensions of privacy, including informational privacy.

HR AI relevance

HR AI frequently processes large quantities of personal information.

For example:

résumés;

employment history;

performance data;

attendance;

behavioural information;

biometric information;

communications; and

potentially sensitive personal information.

An explainability framework should therefore identify:

what data is collected;

why it is collected;

how it is processed;

who can access it;

how long it is retained; and

how employees can challenge inaccurate information.

Key lesson

Explainability is not only about the algorithm; it also concerns the personal data used by the algorithm.

16. K.S. Puttaswamy (Aadhaar) v. Union of India, (2019) 1 SCC 1

Principle

The Supreme Court's Aadhaar judgment addressed questions concerning privacy, proportionality, data use and safeguards.

HR AI relevance

The proportionality framework is particularly useful when examining intrusive workplace AI.

An employer should be able to articulate:

the legitimate purpose;

why the AI/data processing is necessary;

whether a less intrusive method exists;

what safeguards are implemented; and

whether the impact on individuals is proportionate.

For example, using extensive employee behavioural surveillance merely to predict resignation could raise significantly different concerns from using ordinary job-performance information for a promotion assessment.

Key lesson

More data does not automatically mean better or legally justified HR decision-making.

17. Mohinder Singh Gill v. Chief Election Commissioner, (1978) 1 SCC 405

Principle

The Supreme Court strongly emphasised the importance of reasons in administrative decision-making.

A decision-maker generally cannot later substitute entirely new reasons for the reasons supporting the original decision.

HR AI relevance

This has important implications for explainability reports.

Suppose an employee is told:

"You were denied promotion because of your performance score."

During litigation, the employer later says:

"Actually, the AI considered attendance, personality indicators and several other variables."

That creates a serious transparency problem.

The organisation should maintain a contemporaneous record of:

the factors relied upon;

the decision;

the reasons;

the AI output; and

the human review.

Key lesson

The explanation should be contemporaneous and consistent with the actual decision-making process.

18. Kranti Associates (P) Ltd. v. Masood Ahmed Khan, (2010) 9 SCC 496

Principle

The Supreme Court extensively discussed the importance of recording reasons in judicial, quasi-judicial and administrative decision-making.

Reasons provide transparency and enable affected persons and reviewing authorities to understand why a decision was made.

HR AI relevance

This is one of the most useful cases for the concept of an explainability report.

An AI system might produce:

"Risk score: 0.83."

That is not necessarily a meaningful reason.

The employer should be able to translate the result into intelligible reasons, for example:

"The recommendation was primarily based on three documented performance criteria identified in the applicable job policy."

Key lesson

A numerical AI output is not necessarily a legally adequate reason.

19. Siemens Engineering & Manufacturing Co. v. Union of India, (1976) 2 SCC 981

Principle

The Supreme Court stressed the importance of reasoned orders, particularly where administrative decisions affect rights or interests.

Reasons demonstrate that the authority has applied its mind.

HR AI relevance

This principle is highly relevant where AI is used by public authorities or regulated employment decision-makers.

An explainability report should establish that:

the relevant criteria were identified;

irrelevant considerations were excluded;

the AI output was assessed;

the human decision-maker applied independent judgment; and

the final decision was reasoned.

Key lesson

The decision-maker should demonstrate application of mind rather than mechanically adopting an AI output.

20. Summary of the Case-Law Principles

21. What an Actual Explainability Report Could Look Like

A practical HR AI report could have the following structure.

Section 1 — Decision information

Employee/candidate ID

Position

Decision date

Decision-maker

Type of decision

Section 2 — AI system

Model name/version

Purpose

Deployment date

Validation date

Known limitations

Section 3 — Data

Data sources

Relevant variables

Data quality checks

Data correction history

Section 4 — Individual explanation

AI recommendation

Main factors influencing the result

Relative importance of factors

Relevant job criteria

Counterfactual explanation where appropriate

Section 5 — Bias assessment

Disparate outcome analysis

Relevant group comparisons

Proxy-variable analysis

Fairness testing

Section 6 — Human review

Reviewing officer

Evidence reviewed

Whether AI recommendation was accepted

Whether it was modified or rejected

Reasons for modification

Section 7 — Procedural safeguards

Employee notification

Opportunity to challenge

Appeal mechanism

Correction process

Section 8 — Final decision

Final decision

Human reasons

Date

Authorised decision-maker

22. Example

Suppose an AI recruitment system rejects Candidate A.

A weak explanation would be:

AI score: 42/100. Candidate rejected.

A better explanation would be:

The system ranked the candidate below the minimum screening threshold because the candidate's documented experience did not satisfy two mandatory technical requirements specified in the job description. The result was not based on age, gender, ethnicity, disability or other protected characteristics. A recruiter independently reviewed the résumé and confirmed the relevant qualification gap.

An even stronger explanation could add:

If the candidate had demonstrated the required certification, the candidate would have crossed the screening threshold, subject to verification of the remaining requirements.

This provides substantially greater transparency.

23. Explainability Does Not Mean Disclosure of Source Code

An important distinction must be made between:

technical transparency and legal explainability.

An employer does not necessarily have to provide an employee with:

source code;

proprietary algorithms;

confidential model architecture; or

trade secrets.

But the employer may still need to provide a meaningful explanation of:

the purpose of the system;

the significant factors influencing the decision;

the data relied upon;

the role of human decision-makers;

relevant safeguards; and

the reasons for the final decision.

Thus:

"We cannot disclose the algorithm because it is proprietary"

should not automatically end the inquiry.

24. Explainability and Trade Secrets

There can be tension between:

Employer's interest

Protect:

source code;

model architecture;

training methodology;

proprietary datasets;

vendor technology.

Employee's interest

Understand:

why the decision occurred;

whether information was incorrect;

whether discrimination occurred;

whether the decision was arbitrary; and

how the decision can be challenged.

A balanced approach may involve disclosure of decision-relevant factors without disclosure of the underlying proprietary code.

Courts may also use confidentiality protections where appropriate.

25. Explainability Reports as Litigation Evidence

An explainability report can become important evidence in:

discrimination litigation;

wrongful termination claims;

service matters;

recruitment challenges;

promotion disputes;

wage disputes;

privacy litigation;

disciplinary proceedings; and

judicial review.

The report can demonstrate that the employer:

validated the AI;

used legitimate criteria;

checked for discriminatory outcomes;

maintained accurate data;

provided human oversight;

recorded reasons; and

maintained an appeal mechanism.

Conversely, the absence of documentation can create significant evidentiary difficulties.

26. Risks of Poor Explainability

Poorly documented HR AI can create several risks.

1. Discrimination risk

Historical discrimination can be reproduced by the model.

2. Arbitrary decision-making

Employees may receive unexplained scores without rational justification.

3. Privacy violations

Excessive personal information may be processed.

4. Data inaccuracies

Incorrect HR records can produce incorrect AI recommendations.

5. Automation bias

Managers may blindly follow algorithmic recommendations.

6. Accountability gap

The employer may blame the technology vendor while the vendor blames the employer.

7. Litigation risk

The inability to explain the decision can weaken the employer's defence.

27. Best-Practice Legal Framework for Employers

An employer deploying HR AI should ideally adopt an AI Explainability Policy covering:

Before deployment

purpose assessment;

impact assessment;

discrimination testing;

privacy assessment;

data-quality assessment;

validation.

During operation

monitoring;

periodic bias testing;

audit logs;

model-version tracking;

human oversight;

incident reporting.

At the time of an adverse decision

record AI output;

record important factors;

conduct human review;

document reasons;

notify the affected individual where appropriate;

provide correction/appeal mechanisms.

After deployment

periodic revalidation;

audit;

employee feedback;

model retirement where unreliable;

preservation of relevant records.

28. Indian Legal Position — Overall Assessment

India does not presently have one single, comprehensive judicial doctrine specifically titled "explainability of HR AI."

Instead, the legal framework must be constructed from broader principles concerning:

equality;

non-arbitrariness;

natural justice;

reasoned decision-making;

privacy;

proportionality;

administrative accountability; and

employment law.

Consequently, an HR AI explainability report should not be viewed merely as a technical AI document.

It is potentially a legal compliance and evidence document.

29. Conclusion

Explainability reports for HR AI are mechanisms for converting an opaque algorithmic outcome into an understandable, reviewable and legally defensible employment decision.

The strongest approach is not simply:

AI → Score → Decision

but:

Data → AI analysis → Explanation → Human review → Reasons → Final decision → Review/appeal

The Indian Supreme Court's jurisprudence on Article 14, Article 21, non-arbitrariness, natural justice, privacy and reasoned decision-making provides a strong conceptual foundation for demanding greater transparency when AI materially affects employment decisions.

The cases of E.P. Royappa, Maneka Gandhi, Ajay Hasia, Shrilekha Vidyarthi, Puttaswamy, Mohinder Singh Gill, Kranti Associates and Siemens Engineering collectively demonstrate an important principle:

An employer should not be able to transform an otherwise reviewable employment decision into an unexplained decision merely by inserting an algorithm between the decision-maker and the employee.

For HR AI, therefore, explainability is closely connected with fairness, accountability, non-discrimination, privacy and meaningful human oversight.

Available next action: Create a downloadable PDF file here in this chat containing the findings and recommendations above