AI Recruitment Screening in Hong Kong: Law and Bias

Legal limits, bias risk and practical approaches

The use of AI résumé screening tools continues to grow among Hong Kong businesses, and the efficiency advantage is plain: the time to process a large volume of applications drops from days to minutes. But that advantage comes with two issues that need to be faced in Hong Kong’s legal environment: how current anti-discrimination law applies to screening results produced by AI, and the systematic bias already documented in today’s AI tools.

AI Recruitment Under Hong Kong’s Legal Framework

Hong Kong’s Employment Ordinance (Cap. 57) currently has no specific provision on algorithmic hiring decisions. But the Sex Discrimination Ordinance, Disability Discrimination Ordinance, Race Discrimination Ordinance and Family Status Discrimination Ordinance all apply to the recruitment process, including where AI tools are used.

The applicable legal principle is indirect discrimination: if an AI screening system has a disproportionately negative effect on applicants in a protected category, it may still amount to discrimination in the legal sense even when the system design does not explicitly use a protected characteristic as a screening criterion. In learning from historical hiring data, AI may internalise the bias in past decisions: if a type of role was mainly held by a particular group in the historical data, the model may score applicants from other groups lower.

Research by the Hong Kong Legislative Council notes that existing anti-discrimination law may not fully address the complexity of AI-generated bias, and the legal response is still under discussion.

For the same principle applied in insurance underwriting, see Hong Kong Insurance AI Underwriting: Regulatory Framework and Deployment.

Documented Bias Data

Existing research gives concrete data on bias in AI recruitment tools. A 2024 University of Washington study found that in résumé screening, large text embedding models favoured white-associated names in 85.1% of cases, and disadvantaged Black male applicants in up to 100% of test cases.

A study published in May 2025 by the University of Hong Kong and the Chinese Academy of Sciences tested five major LLMs and found that the models systematically scored female candidates differently, and scored Black male candidates lower than white male candidates. These studies target the recruitment scoring scenario directly, and their conclusions are directly relevant to businesses deploying AI recruitment tools in Hong Kong.

PDPO Requirements for Candidate Personal Data

AI recruitment workflows involve collecting and processing large amounts of applicants’ personal data. The PDPO’s requirement on notice at data collection applies: applicants must be told the purpose for which their data is collected. If personal data is used for a new purpose beyond the original collection purpose (for example, using old application data to train an AI model), the data subject’s express consent must be obtained. The PCPD’s Code of Practice on Human Resource Management provides specific guidance in the recruitment context.

Hong Kong Currently Has No Mandatory Disclosure Requirement

Current Hong Kong law contains no express rule that a business must disclose to applicants when it uses AI in recruitment. The PDPO’s collection notice requirement applies, but it is not an AI-specific rule. This differs markedly from some European jurisdictions (for example, EU AI regulation requires transparency disclosure for high-risk AI systems). Until regulation becomes explicit, proactive disclosure is a prudent approach to compliance risk management.

Practical Recommendations for Businesses Adopting AI Recruitment Tools

Conduct bias testing on local actual application data before deploying a model, and do not rely solely on the generic accuracy figures supplied by the tool vendor. Keep a human review stage, especially at the final hire-or-reject decision: the function of AI screening is to narrow the field, and replacing judgement is outside that function. Set up an applicant appeal mechanism so that applicants affected by AI screening have a route to request human review.

For the practical limits of applicants using AI to write their application materials, see AI-Written Cover Letters: Structural Advantage and Personalisation Limits.

Summary

The legal framework for AI résumé screening in Hong Kong is not yet fully defined, but existing anti-discrimination law already provides basic constraints. The documented bias problems are concrete and go beyond the theoretical. Businesses adopting AI recruitment tools need to build a clear institutional design between the efficiency advantage and bias risk management, and avoid assuming that the tool’s “objectivity” automatically solves the bias of human screening. The actual data shows that the two have different bias patterns, and that both exist.