INTRODUCTION
The integration of Artificial Intelligence (AI) into human resource management has changed traditional hiring processes. Modern companies deal with hundreds of thousands of job applications every day, making manual resume screening and candidate evaluation very inefficient. To improve their operations, businesses are using automated hiring tools, algorithmic filters, and predictive assessment models to simplify the recruitment process. These technologies aim for mathematical objectivity, which could help reduce human biases related to gender, ethnicity, and socio-economic background.
However, new technical data and global lawsuits show a different reality. AI recruitment tools often replicate, reinforce, and increase structural bias instead of being neutral gatekeepers. When an automated tool filters out qualified candidates, a complex legal issue arises: who is responsible for algorithmic bias? Since automated systems cannot be prosecuted under current criminal or labour laws, the legal community is left with a significant gap. To address accountability for algorithmic discrimination, it is necessary to examine technical weaknesses, review constitutional requirements, and assess liability within India’s evolving legal framework.
THE MECHANICS OF ALGORITHMIC BIAS IN HIRING
To understand legal accountability, it is essential to dismantle the misconception that AI tools operate in a vacuum of pure logic. AI models learn by identifying patterns within historical data sets. If the historical data utilised to train an algorithm reflects societal prejudices or systematic inequalities, the AI will internalise and reproduce those biases under the guise of objective analysis. This phenomenon is commonly referred to as “garbage in, garbage out”.
Furthermore, algorithms frequently employ “proxy variables” that lead to indirect discrimination. Even if an AI tool is explicitly programmed to ignore protected characteristics such as race, gender, or age, it may utilise correlated data points to achieve the same discriminatory result. For instance, an algorithm tracking residential zip codes, commuting distances, or specific extracurricular activities can inadvertently filter out candidates from specific socioeconomic or ethnic neighbourhoods. The lack of transparency in “black-box” algorithms makes it exceptionally difficult for rejected candidates to identify, let alone prove, that algorithmic discrimination took place.
THE PROBLEM OF LEGAL PERSONHOOD AND ACCOUNTABILITY
Under foundational legal doctrines, accountability requires a recognised legal subject – either a natural person or a juristic entity. Currently, AI Systems possess no independent legal personhood. An algorithm cannot be sued, fined, or held liable under civil or employment laws. Consequently, legal responsibility must be traced back to the human actors involved in the life cycle of the AI: the software developers who designed the system, or the employers who deployed it.
- . However, applying corporate liability to AI decision-making creates significant hurdles. Corporations often assert that autonomous algorithms act as independent intervening factors. Because traditional corporate fault relies on proving human intent or direct negligence, the opaque “black-box” nature of deep-learning algorithms allows corporations to evade liability by claiming the bias was an unpredictable outcome rather than an intentional act.
- Product Liability Frameworks: Treating discriminatory AI as a “defective product” allows claims under product liability torts, such as design defects or failure to warn. While this avoids the need to prove discriminatory intent, product liability regimes is word designed for static tangible goods, not dynamic, self-learning algorithms. Developers routinely shield themselves by classifying AI as software services rather than products, disclaiming warranties, and transferring operational risk to end-users via restrictive licensing agreements. Consequently, standard product liability fails to hold developers accountable for emergent biases that develop post-deployment.
LIABILITY FRAMEWORK: THE EMPLOYERS vs. THE DEVELOPERS
In employment law, the primary burden of ensuring non-discriminatory hiring practices historically rests upon the employer under the doctrine of vicarious liability and statutory duties.
- Product Liability and Tort Law:
Where developers face exposure is through product liability and tort law. If a developer markets an AI recruiting tool as an unbiased, objective mechanism, but the software contains fundamental architectural flaws, claims may arise under Restatement (Third) of Torts: Products Liability for design defects, breach of express warranty, or fraudulent misrepresentation. Internationally, regulatory frameworks like the EU AI Act (2024) explicitly label employment AI as “high-risk”, imposing direct statutory mandates on developers to implement strict data governance, bias testing, and risk management systems before market deployment.
EMERGING REGULATORY RESPONSES: INDIAN AND GLOBAL PERSPECTIVES
- The Indian Landscape
In India, the legal framework governing AI in recruitment remains fragmented. The Constitution of India guarantees the Right to Equality under Articles 14 and 15.
Globally, regulatory approaches are becoming increasingly stringent. This classification mandates that developers and deployers implement rigorous data governance, continuous human oversight, a high level of cybersecurity, and impact assessments before such tools can enter the European market.
Similarly, in the United States, local jurisdictions are implementing targeted laws. New York City’s Local Law 144[5] mandates that employers using automated employment decision tools (AEDTs) must subject those tools to an independent, annual bias audit, the results of which must be publicly disclosed on their websites. Such laws signal a global shift away from reactive litigation towards proactive, mandatory transparency.
CONCLUSION: THE PATH TOWARD ALGORITHMIC FAIRNESS
Algorithms cannot be held independently accountable for discrimination, nor can they be penalised under the current legal system. Ultimate accountability must remain anchored to the human institution that builds, profits from, and deploys these technologies. Allowing employers to shield themselves behind the supposed objectivity of automated tools undermines decades of civil rights and labour law advancements.
fully automated rejections, thereby ensuring that final employment decisions remain bound to direct human oversight and legal responsibilities. Only through these concrete statutory mechanisms can recruitment technology server administrative efficiency without institutionalising automated discrimination.
References:
[1] Jeffery Dastin, ‘Amazon scraps secret AI recruiting tool that showed bias against women’ Reuters (11 October 2018) <https://www.reuters.com/article/world/insight-amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK0AG/> accessed 25 June 2026
[2] The Constitution of India 1950, arts 14 and 15
[3] Information Technology Act 2000
[4] National Strategy for Artificial Intelligence (NITI Aayog, 2018)
[5] New York City Local Law No 144 of 2021
