THE REGULATORY GAP IN THE EMPLOYMENT EQUITY ACT: ALGORITHM HIRING AND INDIRECT DISCRIMINATION IN SOUTH AFRICA

Introduction

Over the years, the growth of automated systems and artificial intelligence in recruitment has changed how employers in South Africa search for and hire employees. With technologies like Applicant Tracking Systems (ATS), applicants now often receive no response at all rather than rejection emails. These systems enhance efficiency and objectivity by reviewing and scanning large volumes of applications against established criteria and filtering out those that do not match the requirements. An ATS functions by scanning candidates’ curricula vitae (CVs) and filtering them before reaching hiring managers. This automation raises critical legal and ethical issues, especially regarding discrimination. The primary legal framework for preventing workplace discrimination in South Africa is the Employment Equity Act (EEA), which aligns with the Constitution to protect individuals from unfair discrimination.¹ This legal article contends that while the EEA prohibits indirect discrimination, there exists a substantial regulatory gap in practice, particularly concerning biased algorithmic hiring. The unique characteristics of automated systems—their complexity and opacity—create significant evidentiary burdens for applicants and pose challenges for the legal system. This makes the EEA’s protection difficult to enforce in this new technological context.

The Framework for Indirect Discrimination Under the EEA

The Employment Equity Act 55 of 1998 aims to achieve fairness in the workplace by promoting equal opportunity and fair treatment through the removal of unfair discrimination.² Section 6(1) of the EEA prohibits employers from unfairly discriminating, whether directly or indirectly, against employees or job applicants based on various factors, including race, gender, pregnancy, marital status, and others.³ While direct discrimination is simple and easy to identify because of its obvious language or actions, indirect discrimination is more subtle. It occurs when a policy, practice, or requirement that appears neutral on its surface has an adverse effect on a particular group protected under the EEA and cannot be justified.

The Constitutional Court established the test for unfair discrimination in Harksen v Lane NO and Others.⁴ This test emphasizes that unfairness is determined by the impact of the conduct on the person. It assesses whether a practice treats individuals differently based on specific grounds and whether such discrimination is unfair. For indirect discrimination claims in employment, applicants must initially present a prima facie case, showing that a particular employment policy or practice adversely affects them and others within a protected group.⁵ Once established, the burden shifts to the employer to demonstrate that the conduct is not unfair or that the policy is justifiable. For example, the employer can provide evidence to show that the decision was based on genuine job requirements.⁶

In Independent Municipal and Allied Trade Union and Another v City of Cape Town, the court ruled that a physical fitness test was indirectly discriminatory as it unfairly excluded women, which the city could not justify.⁷ This case demonstrates that the legal framework clearly challenges practices that appear neutral but produce discriminatory outcomes.

The Rise of AI in Hiring

Artificial intelligence systems are transforming hiring processes. Today, small and large companies use these systems to streamline recruitment instead of manually sorting through numerous CVs. HR departments increasingly adopt these automated tools. Some argue that AI improves fairness by removing human bias and focusing on qualifications rather than appearances. However, this advantage depends entirely on whether the data used to train the AI is fair. South African data often reflects historical inequalities, making this assumption questionable.

Automated hiring tools vary from simple keyword screening software to complex machine learning algorithms that analyze video interviews and predict job performance. These systems aim to treat all candidates equally by applying consistent rules. However, they can still embed and amplify human biases, leading to unfair outcomes. This type of algorithmic discrimination often manifests indirectly.

For instance, when an algorithm is trained on historical hiring data from a company that traditionally hired middle-aged white men, it may learn to link male characteristics with job success, then systematically downrank African female applicants. In feature selection and engineering, the identification of variables for algorithmic decision-making must be carefully scrutinized. Including protected characteristics such as gender or ethnicity risks indirect discrimination unless justified under specific legal frameworks such as affirmative action policies.⁸ This demonstrates why a tool using race, gender, and postal codes as filtering criteria creates indirect discrimination—postal codes serve as a proxy for race due to South Africa’s spatial inequality legacy stemming from apartheid laws.

The algorithm may automatically filter out a candidate from an economically disadvantaged area such as Soweto based on postal code, even when they possess identical qualifications. Despite potentially matching qualifications, the algorithm uses address as a proxy to replicate and reinforce historical socioeconomic disparities. This constitutes indirect discrimination because the neutral criterion (postal code) has a disparate impact based on protected grounds (race and socioeconomic status).

Transparency, POPIA, and the Evidence Problem

The primary challenge that algorithms pose for the EEA framework concerns transparency. Many advanced algorithms operate as a “black box”—so hidden that even their creators and managers may not be able to explain exactly why a particular decision was made.⁹ The decision-making process is concealed within complex mathematical models and vast datasets. For a rejected job applicant, this creates an almost insurmountable challenge. Unlike in traditional recruitment where a human decision-maker can be questioned, in an algorithmic context the “policy or practice” is the algorithm itself, its training data, and its operational parameters. An unsuccessful applicant is unlikely to know that an algorithm was used, let alone have access to the information needed to demonstrate that it impacted them based on a protected ground.

Although no direct law addresses AI hiring specifically, section 71(1) of the Protection of Personal Information Act (POPIA) grants individuals the right not to be subject to decisions made solely through automated processing, and the right to be given reasons for that decision.¹⁰ This means applicants have the right to know the reasoning behind their rejection decision. However, when AI is used, applicants typically receive no explanation for their rejection—only notification. Nonetheless, POPIA’s requirement for reasoning can help a rejected applicant establish a prima facie case of discrimination under the EEA by providing evidence of the algorithm’s impact.

However, POPIA does not provide equality protection equivalent to the EEA. Therefore, POPIA and the EEA must be interpreted together to address algorithmic hiring discrimination. POPIA can help address the evidence problem, while the EEA provides the framework to challenge the unfairness of discrimination in AI hiring systems.

Analyzing the Regulatory Gap in Practice

When the legal test for indirect discrimination is applied to algorithmic hiring, the practical regulatory gap becomes evident. The EEA was written long before the widespread use of AI in the workplace, and it is not equipped to address the specific challenges it presents.

The Evidence Burden

First, the evidence burden on the applicant represents a major obstacle. The EEA places the burden of proof on the person complaining to establish a prima facie case.¹¹ To do so in an algorithmic context, an applicant must provide statistical evidence showing that the hiring tool unfairly disadvantages their group. Because applicants lack access to both the employer’s application data and algorithmic decision-making processes, gathering this statistical evidence is practically impossible. Without such evidence, a prima facie case cannot be established, leaving discrimination allegations unsubstantiated and employers without accountability. This effectively shields algorithmic discrimination from meaningful legal review.

Simultaneously, concerns arise that vague allegations of discrimination, without clear or substantial grounds, could overwhelm government resources allocated to enforcement.¹¹ This tension creates a paradox: the legal framework cannot practically address algorithmic discrimination while remaining protected from frivolous claims.

The Justification Problem

Second, justification becomes problematic. If an applicant manages to establish a prima facie case, the employer would have to justify the practice. An employer might argue that the algorithm simply identifies candidates meeting genuine job requirements. However, if the employer themselves cannot explain the algorithm’s decision-making process, they cannot adequately defend it. Relying on a third-party vendor’s assurance of neutrality is insufficient, as employers remain legally responsible for any discriminatory outcomes under the current framework.¹² The lack of transparency makes it difficult for both employers to meet their legal obligations and for courts to assess whether the justification provided is fair.

The Accountability Gap

Finally, accountability presents a serious challenge. The EEA holds the employer responsible for discriminatory acts. Yet when the discriminatory tool is designed and operated by an outside technology company, it creates a confusing chain of responsibility. While the employer is legally liable, the source of bias lies with the technology provider. The current legal framework provides little guidance on how to allocate liability or compel transparency from a third-party vendor who may claim their algorithms are proprietary business secrets. This leaves applicants without a clear path to redress and leaves employers exposed to legal risk for technology they may not control or fully understand.

Recommendations to Close the Regulatory Gap

1. The Department of Employment and Labour should release a Code of Good Practice on the use of AI in recruitment. This code should require bias testing and maintain records of training data used to develop and deploy algorithmic hiring systems.
2. Section 11 of the EEA should be amended to shift the burden of proof in cases involving automated decision-making. Employers should bear the burden of proving that their algorithms do not discriminate, reversing the current requirement that applicants establish discriminatory impact.
3. The law should establish joint and several liability for both employers and technology providers to ensure accountability and compel transparency, even when algorithms are protected as proprietary information. This dual responsibility would incentivize the development of fair systems and provide applicants with multiple avenues for redress.

Conclusion

Although South Africa’s Employment Equity Act actively seeks to prevent indirect discrimination, it may not effectively address the realities associated with algorithmic hiring. The legal framework relies on identifying a specific policy or practice and requires the complainant to demonstrate its adverse effects. This approach is inadequate when dealing with complex and opaque automated systems, placing an excessively burdensome proof requirement on the applicant. Accountability in these situations is often unclear, making it difficult to justify algorithmic decisions. This regulatory gap can lead to instances of discrimination that may go unnoticed or unchallenged in legal proceedings.

To maintain the effectiveness of the Employment Equity Act in preventing unfair discrimination, adjustments to the law are necessary, or the creation of a specific Code of Good Practice regarding the application of AI in the workplace is needed. Such changes are essential to address issues related to algorithmic transparency, testing, and accountability among employers and technology developers, ensuring that the principles of justice and equality remain upheld in our increasingly automated future.

Bibliography

Primary Sources

Legislation

• The Constitution of the Republic of South Africa, 1996.
• Employment Equity Act 55 of 1998.
• Protection of Personal Information Act 4 of 2013.

Case Law

• Independent Municipal and Allied Trade Union and Another v City of Cape Town [2005] ZALC 10 (LC521/03); [2005] 10 BLLR 1084 (LC); (2005) 26 ILJ 1404 (LC) (18 July 2005).
• Harksen v Lane NO and Others [1997] ZACC 12 (7 October 1997).

Secondary Sources

• Volosevici, D. Artificial Intelligence in Employment Decision-Making: Legal Challenges and Implications (International Academic Publisher, Bucharest, Paris, Calgary, 2025) pp. 34-41.
• Christoph, G. and Peter, L.R. ‘Employment Discrimination Law into the Future’ Stellenbosch Law Review (2018) 29(2) 237-269.
• POISAT, Paul; CULLEN, Margaret and CALITZ, Andre P. ‘Human Resource Managers’ Perceptions on the Impact of AI on the South African Workforce’ SA Journal of Human Resource Management [online] (2024) 22 pp. 3-13. ISSN 2071-078X.

¹ Employment Equity Act 55 of 1998, s. 6(1).

² Employment Equity Act 55 of 1998, Preamble.

³ Employment Equity Act 55 of 1998, s. 6(1).

⁴ Harksen v Lane NO and Others [1997] ZACC 12.

⁵ Employment Equity Act 55 of 1998, s. 11.

⁶ Employment Equity Act 55 of 1998, s. 6(2)(b).

⁷ Independent Municipal and Allied Trade Union and Another v City of Cape Town [2005] ZALC 10.

⁸ Volosevici, D. Artificial Intelligence in Employment Decision-Making: Legal Challenges and Implications (International Academic Publisher, Bucharest, Paris, Calgary, 2025) p. 38.

⁹ ibid., p. 37.

¹⁰ Protection of Personal Information Act 4 of 2013, s. 71(1).

¹¹ Employment Equity Act 55 of 1998, s. 11; and Christoph, G. and Peter, L.R. ‘Employment Discrimination Law into the Future’ Stellenbosch Law Review (2018) 29(2) 237.

¹² Employment Equity Act 55 of 1998, s. 60.