By – Anaaya Wahi

Abstract

As artificial intelligence is restructuring hiring across India’s advancing economy, its adoption risks embedding instead of disrupting deep-rooted gender inequalities prevalent in the labour market. This article scrutinises how AI-driven recruitment tools which are trained on biased historical employment data might systematically disadvantage women thus reinforcing wage disparities and occupational segregation across sectors. Drawing on research published between 2024 and 2026 it argues that algorithmic neutrality is therefore a myth as when machine learning systems encode the past discrimination, they also reproduce it at scale. The metaphor of ‘algorithmic purdah’ showcases this invisible digital barrier which restricts women’s access to economic opportunity. 

Introduction

In 2018, Amazon scrapped an AI recruitment tool after finding it had taught itself to penalise resumes containing the word ‘women’s’ and consistently downgraded graduates of all-women’s colleges. The system had educated to reproduce a decade of biased hiring records. That case is now standard, but the 2026 landscape has moved beyond one cautionary tale.

India stands at a crucial stage as AI adoption in HR is accelerating with 87% of Indian enterprises utilising AI solutions as of December 2025. Into a labour market already marked by structural gender inequality where India’s Female Labour Force Participation Rate(FLFPR) stood at 32.8% in 2024 hiring algorithms emerge as reinforcers of existing bias not neutral disruptors. The question this article raises is whether its adoption will magnify those inequalities or be designed to challenge them not if AI should be used in hiring.

The Rise of Algorithmic Recruitment

AI-powered hiring tools now affect every step of recruitment. AI-driven resume screeners rank candidates before a human eye sees their application. Natural Language Processing systems inspect video interviews, predictive algorithms inspect ‘cultural fit.’ By late 2025, 83% of companies globally used AI to assess resumes and the World Economic Forum predicts over 90% now use AI for preliminary screening. In India, platforms like HackerEarth, Mettl and Naukri’s AI-driven tools are integrated IT and manufacturing while a 2025 McKinsey survey of 1,993 Indian firms found 88% using AI in at least one business function with HR among the rapidly expanding areas. However, efficiency relies on the flawed assumption of neutral data.

The most significant evidence came from Stanford’s Human-Centred AI group that published the largest study of AI hiring algorithms in May 2026. Analysing 4.1 million applications from 156 employers’ researchers found that if the tool selected minority candidates at the same rate as the preferred group roughly 40,000 more applications would have progressed. Notably, the vendor’s audits had found no bias as it had combined results from all employers instead of assessing each role separately. The study concluded that clean audits could mask discrimination.

For gender, a landmark Nature study from October 2025 co-authored by UC Berkeley Haas, Stanford and Oxford found that when ChatGPT generated resumes across 54 occupations using names it repeatedly portrayed women as younger and less skilled despite identical information. A Brookings Study from August 2025 found men’s and women’s names selected at equal rates in 37% of cases across three AI models and nine occupations. This is not accidental. It is the expected output of training systems on a world that has always undervalued women’s work.

Digitizing the ‘Purdah’

‘Purdah’ means literally ‘curtain’ in Persian and Urdu and has traditionally described the segregation of women from society in parts of South Asia. ‘Algorithmic purdah’ describes the same logic digitally. AI hiring tools risk excluding women from economic participation through the hidden logic of statistical inference not direct exclusion.

The dangerous is its opacity. A rejected applicant gets no explanation. An HR manager relying on automated shortlist cannot see why certain profiles ranked lower. When organisations depend on the same vendor’s algorithm what the Stanford study calls an ‘algorithmic monoculture’ biased outcomes increase across industries simultaneously. Proxy bias worsens the issue as algorithms that reward continuous employment penalise women who have taken care breaks, a pattern largely gendered in India. A 2025 study in the Journal of Industrial Relations verified that AI tools trained on biased data are ‘highly susceptible to systematic biases against women in male-dominated fields.’ When algorithms are trained on India’s occupationally segregated system they learn to direct women back to similar sectors. The bias forecloses the future.

The Indian Context

India’s FLFPR of 32.8% is one of the lowest globally for a large economy reflecting deep structural barriers. A November 2025 study using PLFS data found that roughly 90% of India’s female workforce remains informal where gig platform algorithms increasingly control access to work. Ironically, in the sectors using AI hiring most aggressively have women constituting only one-fifth of the IT sector’s labour force and NASSCOM data shows women in IT-BPM leadership fell from 20 to 17 percent even as graduate hiring rises. A NASSCOM study from April 2026 finds women focused on high-risk service jobs like customer support, administration, medical clerking exactly where AI adoption is highest. While 96% of women in India is desired to build AI skills in 2025 the barrier is access not readiness. That is a structural failure. A 2025 investigation found that 40% of AI-driven rejections in India unequally affected women and disadvantaged groups. Reducing India’s gender employment gap the World Bank projects could add billions of dollars to GDP allowing AI to increase it would be a failure of fairness and economic logic.

Regulation and the Path to Inclusive AI

The global regulatory response is advancing. The EU AI Act’s employment rules took full effect in August 2026 classifying hiring AI as ‘high-risk’ and enforcing transparency and human supervision. A 2025 US ruling in Mobley v. Workday ruled that AI tools can be regarded an ‘agent’ of the employer for liability for discrimination a benchmark that should shape global policy. India’s response remains voluntary. The India AI Governance Guidelines from MeitY in November 2025 promote fairness as a key principle but adopt a flexible model without mandatory rules. What is needed is mandatory gender impact assessments before rollout, grievance systems letting candidates challenge AI-driven rejection and investment in datasets that correct for past underrepresentation. Voluntary fairness is easily ignored.

Conclusion

Algorithms are not infallible. They are products of the communities that build them and the data they are trained on. A 2025 Nature study, a Stanford analysis of four million applications a Brookings audit of nine occupations and investigations into Indian platforms all reach the same conclusion that algorithmic bias is the expected outcome when systems trained on unequal histories are used without real accountability.

In a country where gender disparity is as much shaped by social attitudes as economic system, AI tools that blindly reinforce the past become tools of the present’s replication. An invisible barrier that limits women’s access to job opportunities that cannot be seen or challenged by those it excludes is exactly what unregulated AI hiring risks are building quietly and with algorithmic authority. Algorithmic purdah is not an unavoidable outcome. It is a policy choice shaped through the decisions of engineers, firms and policymakers. Technological progress that does not incorporate social justice alongside, is not progress at all.

About the Author

Anaaya Wahi is a fourth-year B.Sc. (Hons.) Economics student at O.P. Jindal Global University. Her interests lie at the intersection of economics, finance, public policy, AI and technology. She is mainly interested in understanding how emerging technologies shape labour markets and social inequalities.