Tanishq Tiwari and Abdul Lateef Khan
I. Introduction
The advent of Artificial Intelligence (AI) and its resultant automation in labour regulation have resulted in a paradigm shift in manual procedures such as hiring, workplace management and task allocation, particularly in sectors such as manufacturing, logistics and gig economy. This automation has bolstered productivity, efficiency and scalability across industries.
The Government of India on November 21st 2025 scrapped the original 29 labour laws and implemented the 4 Labour Codes. This marks a landmark shift aiming to consolidate and strengthen worker protections by ensuring universal wage coverage, formal recognition of gig and platform workers, and streamlined access to social security and grievance redressal, bringing India’s labour regime closer to the realities of a 21st-century workforce. Notwithstanding this, the silence of codes on AI regulated work environment has created a legal vacuum risking potential exploitation of labours.
This article presents a critical analysis of the vacuum regarding AI based regulation in the new labour codes in two parts. Part I identifies critical policy gaps that render the labour codes silent in the face of AI-driven workplace management. Part II suggests policy adoptions taking from best global practices as well as policy recommendations.
II. Critical Policy Gaps against AI-regulated work management
A. The four labour codes fail to address a legal vacuum that lies in the definition of normal working day. The concept of working day in the codes is human-centric ignoring any possibility of AI replacing humans as employers, thus leaving a loophole that treats availability as free labour.
The Code on Wages 2020 (Hereinafter referred to as CoW) under:
- Section 13– defines “normal working day”; and,
- Section 14– mandates rate “not less than twice the rate of wages” for overtime work.
This framework operates under the assumption of identifiable hours of labour performed within a conventional human-centric employment relationship. Algorithmic management relies on Just-in-time-scheduling (JIT) scheduling, which refers to systems when platforms assign tasks only when demand algorithms align, keeping workers logged in for hours but paying only for completed tasks. Within such regulatory atmosphere drivers and delivery workers must stay “logged in” during peak hours to receive orders and are only compensated for hours spent delivering the service and are almost never compensated for the logged in waiting time. Similar tactics are applied by the Urban Company’s auto-acceptance feature which auto-assigns tasks with no refusal right, but pays only for service time, not travel or waiting.
B. Further, the codes operationalize a framework of termination that assume deliberate human action under the definition of retrenchment, rendering algorithmic firings legally invisible and denying workers the procedural safeguards.
The Industrial Relation Code 2020 (Hereinafter referred to as IRC) under:
- Section 2(zh)– defines “retrenchment” as termination by the employer of the service; and,
- Section 70– prescribes the mandatory conditions precedent to retrenchment, which required that prior to retrenchment the employer must provide one month’s notice stating reasons for termination, pay compensation equivalent to fifteen days’ wages for each completed year of service, and notify the appropriate government.
This framework works under the assumption that termination is the product of a deliberate human decision, discounting the integration of AI regulatory tools such as Amazon’s Time-Off-Task (TOT), which tracks workers in real-time. TOT measures workers’ productivity in terms of mathematical efficiency and discounts human hurdles. It identifies workers who may be taking too much time and discounts reasons which may be causing the inefficiency such as injury or illness, and further awards warnings and terminations to these workers without any prior notice or compensation.
C. Additionally, AI has been observed to display gender bias in hiring and pay, yet the codes’ anti-discrimination framework assumes bias must be human and intentional, rendering algorithmic prejudice embedded in training data, not conscious decisions legally invisible.
The CoW under:
The IRC under:
- Section 84 prohibits all unfair labour practices listed under schedule II of the code including arbitrary dismissals, reliance on false or unreliable evidence and disproportionate punishment.
This framework suffers from the recurring assumption that the regulatory body is not a machine that operates purely on calculations and discounts human factors in decision making. AI algorithms are trained using pre-existing data sets which can lead to discrimination against particular demographics. Amazon’s AI-based recruitment tool has been recorded to show bias against hiring women workers due to its training on historical resumes which displayed male dominated field. Ubers’ facial recognition system has shown consistent failure towards darker skinned individuals. In another example, the University of Washington researchers varied names associated with white and Black men and women across over 550 real-world resumes and found the large language models (LLMs) favoured white-associated names 85 per cent of the time, female-associated names only 11 per cent of the time, and never favoured Black male-associated names over white male-associated names. This policy gap lies in the Codes’ failure to recognize algorithmic discrimination as a category of unfair labour practice. The IRC’s Second Schedule requires proof of intent which is impossible when bias is embedded in training data.
D. Additionally, the codes are entirely silent on the vast data ecosystems that power AI-driven management, while India’s privacy laws too actively enable the surveillance. None of the four codes contain provisions regulating employee data surveillance. Moreover, the Digital Personal Data Protection Act (Hereinafter referred to as DPDPA) under:
- Section 7(i)– treats “employment purposes” as a valid exemption for processing personal data without consent.
Modern workplaces harvest vast amounts of employee data location, biometrics, browser histories, even keystrokes to feed these AI systems. In practice, this means an employer can vacuum up any personal information (health, movement, social media) on the ostensible basis of “legitimate interest” in the workplace, with no notice to the worker. At Cognizant, employees must jiggle their mouse every five minutes and type every fifteen or face being signed out, billing reduced, managers summoned. Wipro and Firstsource now do the same. Most egregiously, Urban Company quietly fed beauticians’ photos through AI, digitally altering their faces without consent. Workers who missed an app notification because push notifications weren’t enabled had their photos changed anyway. Those who explicitly objected were ignored.
III. Comparative Best Practices and Policy Recommendations
The four gaps identified above are not inevitable. Across jurisdictions, regulators have begun developing responses that India can adapt. Different jurisdictions have arrived at different answers together they providing a rich reservoir of legislative experiments, enforcement actions, and judicial interpretations from which India can draw as per the following recommendation:
A. Firstly, the unpaid availability gap stems from a single failure: the law’s refusal to recognize that ‘logged in’ is a form of work. France and Belgium have closed this gap not by expanding definitions, but by creating a statutory right to disconnect forcing employers to respect boundaries where India’s law imposes none.
France pioneered the right to disconnect in a 2004 Cour de cassation ruling which established that firing an employee for failing to answer after-hours calls is unlawful. And later pioneered the “right to disconnect” in 2017 through the El Khomri law. It requires companies with 50+ employees to negotiate annual agreements defining when workers can ignore work emails, calls, and messages outside working hours.
Belgium unlike French negotiation-centred model in its 2022 legislation mandates codification of practical modalities for the employee’s application of his right to disconnect outside hourly schedules. This includes guidelines on digital tool use to ensure rest time, leave, and respect to workers private life.
India can borrow from France and Belgium by closing the unpaid availability gap via introduction of a statutory right to disconnect that redefines digital presence as a site of employer control rather than voluntary flexibility. When platforms require workers to log in during peak hours, penalize those who refuse, or use algorithms to rank responsiveness, they exert control over them. A right to disconnect would create a clear legal boundary between paid work time and protected rest periods, which would make platforms either pay for being available or give up their scheduling tools that force compliance. By doing this, the law would make companies consider the economic impact of algorithmic management and clarify what a “working day” means in jobs that involve AI.
B. Secondly, the failure to regulate algorithmic termination comes from a misunderstanding. The law sees firing as something that needs a human to do it. The European Union, or EU, and New York have fixed this by not stretching the term “retrenchment” to include code. Instead, they created new rights: the right to know why an algorithm made a decision, the right to have a human review that decision, and the right to appeal. India’s law, focused on the human aspect, ignores the algorithm completely.
The EU Platform Work Directive 2024 gives platform workers the right to information, human intervention, and the right to appeal algorithmic decisions that affect them. When an algorithm deactivates a worker, the platform must explain the automated decision and allow a person to review it. Workers need to be informed about why the algorithm made its decision and should be able to challenge that reasoning before a human who can change the outcome.
New York City’s Local Law 144 takes a different route. It bans any automated employment decision tool, including those used for hiring and firing, unless it has passed an independent bias audit within the past year. The results of the audit must be made public, and candidates must be informed at least 10 days before the tool is used. India could adopt a model similar to the EU work directive for a procedural fix.
This would mean amending the IRC to acknowledge that algorithmic deactivations give a right to an explanation, human review, and the right to appeal, regardless of whether they qualify as “retrenchment.” The New York model would provide a structural fix, requiring annual independent bias audits for any AI tool used in termination. The results of these audits must be publicly available, and workers should be informed before the tool is used. Combining both approaches would ensure that a worker who is fired by an algorithm receives an individual explanation and has the chance to appeal. The algorithm would also go through annual public audits, helping to catch any systemic bias before it affects the next group of workers.
C. Thirdly, the algorithmic bias stems from the law’s assumption that discrimination requires discriminatory intent. The United States has closed this gap by applying existing anti-discrimination statutes to algorithmic outcomes shifting the focus from why an algorithm decided to what it produced.
The United State (US) landmark case of the Equal Employment Opportunity Commission (EEOC) v. iTutorGroup, Inc. observed a company using an algorithm that automatically rejected female and older applicants. The EEOC sued, and the company settled for $365,000. Establishing that employers cannot shield themselves from liability by outsourcing decisions to algorithms. Beyond iTutorGroup, the EEOC’s 2023 technical guidance documents offer India another toolkit. The 2023 guidance adopted the “four-fifths rule” in context of AI algorithms: if an AI tool’s selection rate for a protected group is less than 80% of the rate for the most-selected group, it creates a presumption of discrimination, shifting the burden to the employer to prove the tool is job related and consistent with business necessity.
Indian Codes can operationalise the EEOC’s outcome-based framework through a two-pronged reform: First, statutory amendments may impose liability where employers deploy automated decision-making tools without meaningful human oversight, thereby preventing the abdication of responsibility through technological delegation. Second, the law may introduce a rebuttable presumption of discrimination where statistically significant disparate impact is demonstrated akin to the four-fifths rule shifting the evidentiary burden onto the employer to establish that the algorithmic tool is job-related and consistent with business necessity. An approach such as this would shift anti-discrimination law away from focus on intent to a more outcome-oriented approach in which discrimination coded within software cannot fall through the cracks of regulation.
D. Finally, while the Indian law is silent regarding workplace data because of its treatment as administrative data, the California Act and Ontario Bill have plugged this loophole by mandating requirements for transparency, proportionality, and examination prior to deployment.
Firstly, California’s Consumer Privacy Act (regulations 2026) requires the performance of risk assessment by the employer prior to making use of automated decision-making technology in relation to decisions on employment, termination, remuneration, or assigning work duties. The purposes of such activities, information gathered, risks to the employees, as well as measures undertaken to reduce such risks should be specified. There should also be an attestation of the accuracy of such assessments signed by an executive officer under penalties of perjury.
Finally, Ontario’s Bill 88 (2022) further offers an administrable fix. It requires every employer with 25 or more employees to publish a written electronic monitoring policy disclosing whether and how they monitor employees, the circumstances of monitoring, and the purposes for which collected information may be used. The policy must be provided to all workers within 30 days of preparation and to new hires within 30 days of joining.
India can achieve an optimal solution to its problem in terms of workplace data governance by implementing a multi-tiered approach that borrows ideas from California and Ontario. First, employers that adopt algorithmic management techniques must be required to submit an annual impact assessment report that details the extent and manner of data collection, purpose of data collection, the proportionality and measures used to address such collection. Meanwhile, platforms must be mandated to publicly disclose their workplace data policies on how they monitor employees’ data and impose legal sanctions on those that fail to do so.
IV. Conclusion
India’s four Labour Codes are a meaningful step towards modernisation of the country’s labour framework. However, a major limitation of these laws is that they do not adequately consider the growing presence of Artificial Intelligence in the management of the workplace. This paper has identified four consequential gaps in the existing legal framework: (1) requiring workers to be available without pay, (2) automated terminations without procedural safeguards, (3) anti-discrimination provisions that do not account for algorithmic bias, and (4) unchecked workplace surveillance through broad exemptions in data protection law. Taken together, these gaps enable AI-driven management systems to operate with little to no meaningful legal accountability and the least protection for those who are most exposed to such systems: gig and platform workers.
The reforms suggested in this paper do not entail a complete overhaul of India’s labour architecture. They draw on the legislative experience in France, Belgium, the European Union, the United States, California and Ontario to provide practical and focused interventions: a statutory right to disconnect, procedural safeguards for algorithmic deactivations, outcome-based anti-discrimination standards and stronger obligations regarding workplace data governance. These measures would make the Labour Codes more relevant to the realities of modern-day workplaces. And a legal framework that does not recognise algorithmic management cannot meaningfully protect the workers it governs. Thus, the update of India’s Labour Codes to meet the challenges posed by Artificial Intelligence is not just a matter of legislative modernisation it is an absolute necessity to protect worker dignity in an era of increasingly invisible management.
Tanishq Tiwari and Abdul Lateef Khan are students at Hidayatullah National Law University(HNLU), Raipur.
NUALS Law Journal 