The Singapore Law Gazette

Science Fiction, Real Liability: Legal Risks of AI in HR

Abstract

The growing use of artificial intelligence (“AI”) in human resources (“HR”) presents new legal challenges for employers in Singapore. While AI enhances efficiency and consistency, its role in recruitment, disciplinary decisions, and dismissal raises concerns under existing employment laws. Further, there are fears of additional risks under the forthcoming Workplace Fairness Act 2025 (“WFA”) if the AI makes decisions based on protected characteristics. AI‑driven decision‑making, coupled with emerging deepfake risks, heightens issues of transparency, fairness, and accountability in HR decisions. In this article, we will make the case that these developments require employers to ensure meaningful human oversight, transparency and robust governance protocols in order to ensure that a “human-in-the-loop” model is in place to ensure that AI’s role is only the decision enabler and humans remain the decision makers.

Across decades of science fiction, from classic films to modern streaming series, audiences have been captivated by stories in which humanity’s greatest technological breakthroughs spiral beyond our control. A recurring theme runs through these narratives: humans create an extraordinary innovation, lose oversight and control of it, and unintended consequences follow. While fictional, these cautionary tales highlight the risks of unchecked technological advancement. Today, they resonate strongly with contemporary challenges surrounding the rapid adoption of AI in the modern workplace.

The Growing Use of AI in HR-related decisions

In recent years, employers in Singapore have increasingly deployed AI‑enabled tools to aid decision-making in HR matters. These tools are now commonly used to evaluate employee productivity, identify performance risks and, in some instances, recommend disciplinary action or dismissal. While AI can support more data‑driven decision‑making, its use in employment contexts raises fundamental legal questions about fairness, transparency and accountability. Particularly, where AI outputs influence or determines decisions affecting an individual’s livelihood.¹See generally Ministry of Manpower, Tripartite Guidelines on Fair Employment Practices (as updated).

This tension between technological efficiency and accountability for AI-aided decisions sits uneasily with Singapore’s employment law framework, which has consistently emphasised merit‑based and non‑discriminatory workplace practices. Singapore has further strengthened its measures against discrimination with the introduction of the WFA which is Singapore’s first anti-discrimination law. The WFA protects employees from discriminatory employment decisions throughout the employment lifecycle including hiring, appraisal and termination. The WFA now makes it a statutory tort for employers to make discriminatory employment decisions based on 11 protected characteristics – age, nationality, sex, marital or pregnancy status, caregiving responsibilities, race, religion, language, disability and mental health conditions.²Workplace Fairness Act 2025 (Act 8 of 2025), sections 30 read with 34; See also Ministry of Manpower, “Passing of Workplace Fairness Bill Marks Next Step in Building Fair and Harmonious Workplaces” (6 January 2025), available at: https://www.mom.gov.sg/newsroom/press-releases/2025/passing-of-workplace-fairness-bill-marks-next-step-in-building-fair-and-harmonious-workplaces (accessed 20 February 2026). The WFA makes discriminatory employment decisions a statutory tort by (i) defining discrimination as a prohibited adverse employment decision based on protected characteristics, and (ii) expressly conferring on affected individuals a right to bring a civil action for a “statutory tort of discrimination”, thereby converting breach of the statutory prohibition into a privately enforceable civil wrong with tort-like remedies.

Notwithstanding the increasing prevalence of AI in employment-related decisions, the WFA does not make a distinction between decisions made wholly by humans and those influenced by technological tools. Further, as of the date of writing, neither the Ministry of Manpower (“MOM”) nor the Tripartite Alliance for Fair & Progressive Employment Practices (“TAFEP”) has provided any guidance on best practices. Nevertheless, the practical reality is that decisions informed by AI remain decisions made by the employer. The deployment of AI does not diminish an employer’s responsibility to ensure that outcomes are fair, defensible and compliant with applicable legal standards.

Lesson 1: Overreliance on AI and the Need for Human Oversight

A recurring concern is the risk of accepting AI‑generated outcomes at face value, with employees or managers treating algorithmic scores as objective or determinative in employment decisions. This danger is illustrated in 2001: A Space Odyssey, where HAL 9000, the spacecraft’s intelligent computer, was trusted to control key spacecraft operations. The astronauts’ unquestioning reliance on an ostensibly “infallible” machine and their abdication of the supervisory role over HAL 9000 led to grave errors. The same principle applies in the employment context. If an employee challenges a dismissal on the basis of poor performance, the employer may be required to demonstrate that the decision was fair and substantiated. Where an employer acknowledges that an AI tool contributed to the termination but is unable to explain how the system evaluated performance, the employer may be vulnerable to allegations of unfair dismissal due to the employer failing to exercise independent judgment or provide due process. In such circumstances, overreliance on AI not only undermines the integrity of the decision-making process, but may also weaken the employer’s position in any subsequent legal scrutiny

Emerging best practice that addresses the risk of overreliance supports a “human‑in‑the‑loop” model, under which AI serves as a decision‑support tool rather than a decision‑maker.³Personal Data Protection Commission (Singapore), Model AI Governance Framework 2.0 (2024), available at: https://www.pdpc.gov.sg/-/media/files/pdpc/pdf-files/resource-for organisation/ai/sgmodelaigovframework2.pdf. With legal accountability increasing, employers must ensure human oversight remains for any AI‑assisted process, reinforcing that technology cannot replace the need for sound human judgment. Employment decisions, particularly those involving disciplinary action or dismissal, should be reviewed by an individual with sufficient authority to question and depart from the AI’s recommendation. The review must be holistic , taking into account factors that algorithms may not capture, such as temporary changes in workload, health issues, or qualitative contributions to team performance. This need for human oversight is especially acute in cases of summary dismissal, which carries a higher risk of legal challenge.

Lesson 2: Transparency as the Cornerstone of Accountable AI Use

Closely linked to human oversight is the requirement for transparency. The example of HAL 9000 in 2001: A Space Odyssey—a system that makes decisions the human crew cannot understand, leading its behaviour to appear irrational and unpredictable—illustrates the risks of relying on an opaque technology. Human oversight is meaningful only when decision‑makers can understand, interrogate, and if necessary, challenge the basis of an AI system’s outputs, which is why transparency is its necessary counterpart. To avoid similar uncertainty in the workplace, transparency must operate in two directions. Employers need sufficient visibility from AI service providers to understand what data the system uses, how it interprets the data, and what safeguards are built into the model. In turn, employers should be able to communicate, at an appropriate level, what categories of data were considered, how performance metrics relate to job functions, and what measures exist to mitigate biased outcomes. While such transparency is important from a compliance perspective, it is also a means of maintaining and prioritising trust within the employment relationship where AI is used in determining employment decisions.

Emerging Risks From Deepfakes

Beyond employment decisions influenced by AI, recent developments in AI technology have further expanded the scope of workplace risk. Deepfakes, which are AI‑generated audio, images or videos that convincingly replicate a real person’s appearance or voice, have emerged as a significant concern for employers. A useful illustration of the danger posed by such Deepfakes have been seen in the Mission: Impossible movie series, where hyper‑realistic face swapping masks and near perfect voice mimicking devices allow characters to impersonate others so convincingly that even close colleagues and high‑security systems are fooled. Yet, what was once fictional has now become a real workplace danger, as digital deepfakes can easily infiltrate environments that depend heavily on trust authentication, or recorded evidence.

In the employment context, deepfake technology has been used to impersonate senior executives, deceive finance personnel into making unauthorised payments, and extract confidential information through fabricated urgency.⁴Jonathan Isaacs, How deepfakes and AI are reshaping workplace risk: What HR leaders need to know, Human Resources Online (1 December 2025). These incidents create immediate operational, governance, and legal consequences. When employees act on deepfake instructions, organisations may incur significant financial losses and face scrutiny for inadequate verification controls. For HR practitioners, this underscores the importance of training and awareness programmes to help employees identify suspicious communications and respond appropriately. Deepfakes also heighten workplace harassment risks. AI‑generated images or videos depicting employees in compromising situations can be circulated internally or externally, causing serious personal and reputational harm. Such conduct may amount to workplace harassment and potentially expose employers to vicarious liability where reasonable preventive measures were not taken.⁵See Tripartite Advisory on Managing Workplace Harassment; cf. Protection from Harassment Act 2014.

Data protection obligations compound these concerns. Under the Personal Data Protection Act 2012 (“PDPA”), employers must implement reasonable security arrangements to prevent unauthorised access, use, or disclosure of personal data. When deepfakes are created using employee images, voice recordings, or other employer‑held data, the occurrence of such a breach may itself give rise to an inference that adequate safeguards were not in place, exposing the employer to regulatory liability.

Taken together, these developments demonstrate that AI‑related workplace risk now extends well beyond algorithmic bias or automated decision‑making. It encompasses fraud, harassment, data protection obligations, and the integrity of evidence within the workplace. Addressing these risks requires a coordinated approach that integrates HR policy, legal compliance, and organisational culture to ensure that safeguards are robust, anticipatory, and fit for an evolving technological landscape.

Strengthening Governance: A Unified Approach to AI Risk

Employers are advised to implement internal AI governance frameworks that define acceptable uses of AI in HR functions and,specify the required levels of human oversight, especially for employment decisions. This is even more pertinent in light of the new WFA which imposes statutory penalties on employers for discriminatory employment decisions, and establishes clear escalation and appeal mechanisms for affected employees. Policies should explicitly prohibit the unauthorised use of employee images, voice or personal data, including deepfake creation. Training programmes should equip employees to recognise suspicious requests and encourage reporting without fear of repercussion. Verification protocols, such as dual authorisation and call‑back procedures, can provide additional layers of protection against misuse.

Comparative developments suggest that heightened expectations around AI governance and accountability are becoming the norm. Singapore’s forthcoming WFA appears aligned with this broader international trend, signalling a shift towards clearer expectations of procedural fairness even in technologically assisted decision‑making.

Science fiction’s enduring warning that technology without oversight can produce unintended harm resonates strongly as AI becomes embedded in HR decision‑making. Under Singapore’s WFA, employers remain fully accountable for ensuring that AI‑influenced employment decisions are fair, transparent, and non‑discriminatory. Emerging forms of AI misuse, such as deepfakes, further demonstrate how AI can facilitate deception and harassment, underscoring the need for strong governance, verification protocols, and employee training. Ultimately, AI can enhance efficiency, but only when used responsibly. In decisions affecting dignity, trust, and livelihood, human judgment must remain central, with AI serving as a support rather than a substitute.