abstract. Increasingly, automated processes—under the catch-all term “artificial intelligence” (AI)—serve as “mechanical managers” in the workplace. They may manifest as productivity applications to spur workers to work faster or as deputized surveillants who monitor workers’ every move. Moving beyond surveillance capitalism, this Essay argues that, absent legal intervention, we are on a path toward a scientific approach to management that prioritizes efficiency and deploys AI technologies to maximize output through the collection and exploitation of worker’s captured capital. That is, the more benevolent tenets of scientific management, such as encouraging productivity or achieving mutual prosperity for employers and workers, no longer represent paramount goals for firms. Rather, emboldened by new AI capabilities, firms have set out to quantify or reduce all elements of workers’ experience to data. This data is valuable capital that (1) holds exchange value and (2) drives the automation of workplaces and the displacement of workers.

The contributions of this Essay are threefold. First, this Essay names and describes the sociolegal phenomenon of “captured capital”—that is, the coercive collection and use of worker data to facilitate workplace automation and ultimately worker displacement. Second, this Essay situates this phenomenon within an AI arms race in the workplace and analyzes it through the lens of law and political economy. Specifically, the Essay argues that the AI arms race has spurred the unchecked development and deployment of AI technologies and that a laissez-faire approach to globalization has encouraged the growth of a borderless labor market without adequate international labor protections, leaving workers vulnerable. Third, the Essay sets forth three potential legal avenues for redress: (1) treating the data gathered from workers as stake capital in the automation of their workplaces such that a portion of the gains from automation is rightfully returned to the worker; (2) creating a licensing regime for workers to license their data freely to firms; and (3) requiring firms that use workers’ data as part of their automation process to pay into a fund that will finance a guaranteed income. Finally, the Essay notes a role for the International Labor Organization to play in protecting workers in the AI revolution.

Introduction

A Time article published in January of 2023 revealed that Kenyan workers had played a crucial, yet hidden, role in bringing about the AI technology ChatGPT.1 The workers had received as little as $1.32 per hour for their arduous work labeling the noxious content that would become part of ChatGPT’s training data.2 To produce AI models capable of speech like ChatGPT, words and sentences are scraped off the internet, and that language is used to train models enabling AI to produce speech.3 Kenyan workers were responsible for going through examples of speech and labeling them as racist, sexist, or harmful in some other way.4 The labels that these workers placed on the examples of speech were then fed to ChatGPT, training the AI model to identify (and avoid producing) harmful content in the future.5

This work is crucial to the continued efficacy of ChatGPT. ChatGPT, an AI-powered chatbot, is the brainchild of the corporation OpenAI.6 As of early 2024, the firm was valued at eighty billion dollars.7 Much of that capital has been concentrated at the top, with little recompense flowing to the Global South workers who are the “draught horses” of the AI revolution.8 This business model is not limited to OpenAI. Sama, the San Francisco-based outsourcing firm that OpenAI used in Kenya, has also engaged African workers for other tech companies such as Google, Meta, and Microsoft in their development of AI technologies.9

But this Essay is not only about the exploitation of Kenyan workers. It is also not merely about how such extractive AI development echoes the oil speculation of the twentieth century, with its pell-mell pace and little regard for the human-rights abuses and environmental devastation left in its wake.10 Rather, this Essay is meant to illuminate a danger to workers everywhere: worker exploitation is found not only in the development of AI technologies, but also in their deployment in the workplace.11

Consider the issues at the heart of the recent Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) union strike by actors in the United States. Some of the union’s demands centered on ensuring that AI technologies would not be deployed to capture the actors’ past or current performances for the sole enrichment of studios.12 The union was also concerned that AI technologies would be deployed to eliminate jobs for background actors.13

One actor shared her experience with AI in the film industry when, after a few days of filming, she was directed to report to a trailer for body scanning.14 She recalled being placed in front of “a series of cameras,” where she was instructed: “Have your hands out. Have your hands in. Look this way. Look that way. Let us see your scared face. Let us see your surprised face.”15 A digital replica of her body was created, and she was not told if or how that replica would be used.16 The actor lamented: “I fear that AI is eventually going to weed out background actors. They won’t have any use for us anymore.”17 This actor’s experience is not unique; other actors have reported feeling coerced to submit to such scans.18

This Essay proposes that both the Kenyan workers and the Hollywood actors are primed to be victims of the data revolution wrought by AI technologies.19 And the danger extends further. Workers everywhere are at risk of having their capital captured by corporate firms in the form of data and receiving little to no recompense. Laissez-faire attitudes towards both the development and the deployment of AI technologies in the workplace have introduced new vulnerabilities for workers which demand legal attention.20

Although the Kenyan workers’ dire circumstances and the actors’ union strike have only recently brought the phenomenon of captured capital21 to the public consciousness, worker data has long fueled the development of AI technologies.22 As part of the AI revolution, the capture of workers’ capital may generate profits for the firm in two ways: (1) the data may be leveraged to develop technologies that will eventually displace workers, and (2) the aggregated data may simply be sold to data brokers such as credit-card issuers, pharmaceutical manufacturers, insurance providers, and airlines.23 As an example of the first strategy, Enron emails collected as part of the investigation of the financial scandal in 2003 have widely been used as training data for language-learning models.24 It was from “[u]sing these emails as raw data for real person-to-person conversations” that AI systems such as Apple’s Siri and Google’s Smart Compose—predecessors to more sophisticated systems like ChatGPT—learned how to understand and speak to users.25 As an example of the second strategy, some firms that have profited from selling consumer data have started doing the same with their workers’ data.26 This strategy can be understood as a “form of rent-seeking—in which companies generate huge profits by packaging and selling worker data in [a] marketplace hidden from workers’ eyes.”27

This Essay contributes to law-and-political-economy (LPE) scholarship by tackling head-on the challenges that the AI revolution poses to worker power and self-determination, particularly with respect to workers’ control over data.28 As its foundation, this Essay takes the normative stance that the development and deployment of AI technologies should not come at the price of worker welfare and autonomy. In her prescient book, In the Age of the Smart Machine, business scholar Shoshana Zuboff described two potential paths for technological innovation: informating and automating.29 As Zuboff described it, informating would create technologies that liberate humans from menial work.30 Automating, on the other hand, would augment managerial control through panoptic surveillance, and it would diminish autonomy and dignity at work.31

Currently, we are experiencing the sociotechnical phenomenon of automating. Absent legal intervention, we are set on a path towards an extreme iteration of Taylorism—the theory of scientific management that utilizes scientific methods to increase efficiency in the workplace—where AI technologies are deployed to capture workers’ capital.32 In this extreme, antiworker version of Taylorism, key benevolent tenets of scientific management, such as mutual prosperity for managers and workers alike, would no longer hold as paramount goals for firms.33 Rather, with profit as the principal goal, firms would set out to quantify all elements of a worker’s experience (job efficiency, health, mental state, social behavior, and so on) through the collection of data.34 This evolution of Taylorism is hostile to workers because unlike consumers who often receive some measure of value for the collection of their data (e.g., in the form of reduced prices or personalized services), workers are “only ever compensated for their services—never their data.”35 Furthermore, this worker data is valuable capital that drives the automation of workplaces and that may eventually animate the robots that will displace the human workers who generated the data in the first place.36

This Essay develops a theory of worker data as “captured capital,” analyzes captured capital through an LPE lens, and proposes potential avenues to justice for workers. The focus of this Essay is on the property rights of workers in the data they produce in the workplace. Elsewhere, I have written extensively about the privacy37 and discrimination38 issues associated with the collection of employee data. As others have noted, “[P]rivacy law serves to protect information by punishing those who collect, use, or disclose information without legal authorization or justification,” whereas property law accords the owner of the information “a bundle of potential rights over the property.”39 This Essay is squarely preoccupied with workers’ control of their data.

Furthermore, this Essay is not focused on parsing the distinction between personal information and other business data, as some legal scholars have explored.40 I concede (and have argued elsewhere) that identifiable personal information deserves heightened legal protection.41 However, with the continued development of AI technologies, a sharp distinction between personal and nonpersonal data may no longer exist.42 An underlying assumption of this Essay is that collection of worker data in the realm of employment is a longstanding practice that technology is making easier, more affordable, and even inevitable in many industries.43 Operating on this assumption, this Essay focuses on a pressing legal concern: how control of the data collected in the workplace and the profit accrued from such data should be apportioned between firms and workers.

As a final caveat, this Essay does not touch upon how to calculate the exact value or percentage of capital that should accrue to each individual worker or different types of workers. This is in recognition of wide discrepancies in worker data collection across employment sectors and even across firms. The Essay also acknowledges, and reserves for future discussion, that the issue of calculating captured capital reveals a legal tension: U.S. law mostly values property individually, whereas the captured capital from workers may have the most value in the aggregate rather than the individual form.44

This Essay proceeds as follows. Part I lays out the theory of captured capital and posits legal arguments for why and how both ownership and control of such capital could accrue to workers. Part II situates captured capital in an LPE framework and analyzes the legal, political, and economic conditions of capital capture. In particular, it argues that an AI arms race, the rise of a borderless labor market, and the lack of international labor-law protections concurrently create ripe conditions for the capture of workers’ capital. Part III details three potential avenues for legal redress: an equity-stake approach, a licensing regime, and a guaranteed-income program for displaced workers.

I. a theory of captured capital

The Legal Information Institute defines “capital” as “any asset used for a productive purpose.”45 Such assets may “include tangible items, such as cash or machinery, or intangible items, such as intellectual property or human capital.”46 In prior writing, I have defined “captured capital” as “the data that is siphoned from workers both knowingly and unknowingly as part of the employment bargain.”47 What defines this data as “captured” is “the element of coercion in how it is obtained.”48 What defines this data as “capital” is that it holds “both inherent and exchange value,”49 meaning that either the use or the sale of the data will be profitable for the firm. Captured capital holds inherent value for the firm because it may provide organizational insights that drive greater productivity and efficiency.50 Beyond improving everyday organizational functions, “data gained from the work habits and practical work innovation of workers . . . may even serve to power the automation of jobs.”51 Furthermore, captured capital holds exchange value for the firm because it can be sold (usually in aggregate form) to data brokers, who may then distribute it for uses orthogonal to any actual or fictive worker consent.52

In this Part, I expound on the theory of captured capital. First, I address the question whether this capital is indeed “captured.” I argue that the data rightfully belongs to the worker from whom it has been collected. Second, beyond the question of ownership, I address the question of control—that is, regardless of who owns the capital, who may control the capital and to what extent? I argue that the worker’s initial acquiescence to the employment bargain does not operate as an automatic waiver or affirmative relinquishment of any rights to control the capital.

A. Ownership and Control

The theory of captured capital turns on the premise that workers have ownership rights to their data, which finds support in natural law and statutorylaw.53 Beyond this claim about ownership, this Essay draws from common law to make another central normative claim: that workers should legally retain some control over their data in the workplace, regardless of how ownership of that data is parsed.

First, a Lockean approach to the natural law of property would accord workers property rights to the data created by their labor.54 John Locke argues that any time an individual invests their labor to create something, that creation becomes the individual’s property.55 Workers’ labor—whether it be the writing of emails, the tagging of objectionable content, or the acting, posing, and gesturing performed for body scanning—is the essential element that generates valuable data. Thus, following the Lockean view of property, workers should hold a property claim to that data. Admittedly, some of the employers’ investment is entangled in the workers’ investment: for example, employers may have bought equipment for body scanning or the computer on which the emails are written. But ultimately, the indispensable element of the final output is workers’ labor. The employer might provide the space and resources, but without workers’ labor, the captured capital—that is, the data that holds inherent and exchange value—would not come to exist.

Although the Lockean approach ostensibly champions the rights of those who labor, it has led in practice to the inequitable recognition of certain types of labor as more valuable than others.56 A Lockean approach would also dictate a quantification of each worker’s labor to determine its exact value, as it theorizes that individuals hold ownership over what their own labor creates.57 This would potentially undermine worker solidarity and create valuation issues, as workers’ data is in fact inherently social and communal and not only attributable to a single individual.58 Thus, a Lockean approach ultimately falls short. But other tenets in statutory law also support workers’ ownership of their data.

Recent statutory developments could imply an ownership claim by workers to the data they generate in the workplace. Or, at the very least, those new state laws assert some measure of worker control over their workplace data. The California Consumer Privacy Act (CCPA) covers workers of for-profit companies operating in California.59 The California Privacy Rights Act (CPRA) amended the CCPA to include some previously unregulated businesses, such as non-consumer-facing businesses, and required them to become compliant with the CCPA.60 Under the CPRA, workers have rights regarding their personal data, including the right to be informed when employers collect their data; the right to access the data that has been collected from them; the right to request corrections or deletions to their data; the right to opt out of the sale or sharing of their data; and the right to limit the use of sensitive information.61 The right to opt out of sale of the data implies some ownership rights. Additionally, employers are required to provide privacy notices to employees and job applicants that specify the types of sensitive data collected, whether this data will be sold or shared, and how long it will be retained.62 Through the CCPA and the CPRA, consumers and employees alike in California have explicitly been granted significant rights to their data.

The premise that workers should have some control of the data they generate is also supported by common law. Professor Jack M. Balkin has popularized the concept of “information fiduciaries” as derived from the common-law doctrine of “fiduciary.”63 Balkin uses the term “information fiduciaries” to refer to an individual or entity who has traditional fiduciaries duties (i.e., the duties of care and loyalty) in managing another individual’s asset, which in this case is the individual’s information (e.g., personal and sensitive information and intellectual property).64 This concept is a response to an unavoidable modern dilemma: consumers must surrender large swaths of their data, much of it personal and sensitive, in order to participate in the digital public square and marketplace.65 Such data can then be exploited or even used against the consumers in the practice of surveillance capitalism.66 Consumers are thus pressured into relinquishing personal data for convenient access to the digital world. Workers face a similar but arguably even greater pressure: as part of the employment bargain, many workers must divulge personal information for the mere chance to earn a livelihood.67 Worse still, if they gain employment, workers come under the unremitting surveillance of their employers, who have carte blanche to collect any and all worker data.68 Given the vast amount of data collected from workers, legal scholar Matthew Bodie has argued that a legal designation of employers as fiduciaries, entrusted with workers’ data and obligated by law to use that data in ways dictated by the employee, is appropriate.69 This designation would allow workers to retain some control of their data even if they must relinquish said data to their employers.

B. Overcoming Critiques of the Captured Capital Theory

Intellectual-property law has been presented as a steadfast argument against workers’ control of their data. But this Essay posits that another body of law, corporate law, provides support for workers’ data rights. Corporate law, specifically corporate-governance theories of stake capital, supports workers’ property rights to the data they generate as investments they have made in the firm.70

As Professor Amy Kapczynski has noted, “Intellectual-property scholars have, for the most part, argued vociferously against any form of property protection in personal data for a variety of reasons.”71 And it is true that intellectual-property law privileges innovation and thus would elevate the unfettered flow of data as the law’s paramount goal.72 Take, for instance, two doctrines of intellectual-property law: “work for hire”73 and “the implied duty to assign.”74 The 1909 Copyright Act promulgated the “work for hire” doctrine, which was narrowed by the 1976 Act defining “work made for hire” as “a work prepared by an employee within the scope of his or her employment.”75 More importantly, the Copyright Act of 1976 amended the “work for hire” doctrine to make the employer the author of any work made for hire unless expressly agreed otherwise, by default granting employers intellectual-property rights to workers’ work.76

Opponents to the “captured capital” theory could seize on the “work for hire” doctrine as the foundation for the presumption that all data generated by workers in the workplace belong to the employer who has hired them. Similarly, the “implied duty to assign” doctrine for patent rights could be marshaled as an argument against the theory of captured capital. Although this doctrine is implied, not statutory like the “work for hire” doctrine, legal scholars have noted that “[c]ourts have tended to recognize such an implied duty to assign patent rights in situations where an employee hired to solve a problem engages in research, and the invention relates to that effort.”77

These potential attacks against the theory of captured capital will be unsuccessful because they fail to recognize a crucial distinction: the aforementioned intellectual-property doctrines pertain to the finished product, not to parts or factors of the production process.78 The product/process distinction is important because intellectual-property law is meant to reward the realization of an idea, not merely the idea itself.79 Thus, intellectual-property law generally covers the realized idea or product and not just the process.80

Furthermore, the notion of captured capital draws support from another area of law—corporate law. Legal scholars have previously recognized human and intellectual capital as “factors of production” separate from the finished product.81 This definition supports the idea that workers, through their production of data, are generating an input that employers can then exploit for profit. It also highlights the enduring nature of that input—data that is reusable for employers’ future productive endeavors. The worker-created data is thus a “factor of production” in whatever employers produce later, including AI technologies. Without workers’ investment of their human and intellectual capital and their generation of valuable data inputs, the employers’ finished, market-ready products would not come into existence. Therefore, in analyzing captured capital, one should view employees through the lens of corporate law, in particular corporate governance, and understand that they are stakeholders who are providing necessary “stake”—their data—to firms.82 In providing this data to employers, employees are neither extinguishing their ownership rights nor ceding all control of their data to employers.

II. the law and political economy of captured capital

The LPE movement argues that questions of market efficiency have tended to obscure questions of economic power.83 Governments have allowed power derived from economic superiority to dominate, and law as a process for making society more just has left “market power” largely untouched.84 Thus, a central precept of the LPE movement is that governmental neutrality (i.e., laissez-faire capitalism) has failed to address power imbalances and the inequality that they breed. A central tenet of the LPE approach is that economic activity ought to be accountable to the democratic government that allows it to occur.85 For the LPE movement, the present inequality stems from a false presumption that markets are competitive enough to self-adjust, and as such, that legal inquiries into power dynamics no longer matter.86 The LPE movement champions a hands-on approach to the economy and a move away from the notion of “autonomous” economic ordering.87

The sociolegal problem of “captured capital” stems exactly from a laissez-faire approach to the economic ordering of firms that has enabled them to optimize their firm structures for AI development and deployment without regard for workers’ rights. Thus, an LPE approach calls for governmental intervention to reorient AI policies and practices toward worker equity. In the following Sections, I detail how an AI arms race and inadequate international labor protections have shaped the law and political economy of AI and enabled the capture of workers’ capital.

A. An AI Arms Race

The so-called AI arms race is the idea that states are in geopolitical competition to achieve superiority through the development and deployment of AI. Numerous governments have launched national AI initiatives, with China aiming to be the global AI leader by 2030 and the United States introducing the American AI Initiative and a corresponding defense strategy in 2019.88 The AI-arms-race theory presupposes that developing efficient and effective AI will allow a country (or a business within a country) to achieve greater economic dominance in the national and international markets.89 For example, President Biden recently touted the establishment of an AI data center in Wisconsin as a win for developing stronger American business and improving AI.90

Critics of the AI arms race note that the lack of intergovernmental cooperative policies promotes a warlike, adversarial attitude toward AI development and is likely to result in the irresponsible development of AI.91 Another consequence of the AI arms race is that workers become mere fodder for the automation industry.92 In the Sections below, I discuss how the rush toward AI technologies like ChatGPT and productivity-tracking technologies enables worker displacement.

1. ChatGPT and the End of White-Collar Work?

In February 2023, ResumeBuilder.com surveyed 1,000 U.S. business leaders to see how many companies currently use or plan to use ChatGPT. The findings were as follows: 49% of companies currently use ChatGPT, 30% plan to use the technology, and 48% of companies already using ChatGPT have displaced workers.93 About 25% of companies deploying ChatGPT claim to have already saved at least $75,000, and 93% of current users say they plan to expand their use of ChatGPT.94

One of the most surprising discoveries about ChatGPT has been its ability to generate code.95 Consequently, coders have expressed concerns about job loss.96 Unlike previous automation—which targeted “hard, dirty, repetitive jobs”—this wave of AI innovation is affecting creative and well-educated professionals.97 Goldman Sachs has projected that AI could automate 18% of jobs globally, posing a higher risk to white-collar workers, such as lawyers, than to those in construction or maintenance.98

2. Data Collection as Part of Productivity Tracking

The collection of worker data has reached a fever pitch with the introduction of AI technologies.99 Employers now have access to a wide array of devices that can not only monitor worker productivity but also track workers’ every move.100 Although employers claim that worker tracking is merely aimed at improving productivity and efficiency, the hidden truth is that worker data is already serving as the training data to automate jobs and displace workers. Consider that in 2018, Amazon acquired a patent for a wristband that can detect motions and positions.101 The bracelet has the capability to monitor and direct the worker to the correct inventory bins via haptic feedback.102 Other companies have followed suit with patents for similar wearables, such as gloves or wristbands for workers.103 The motivation for this type of wearable technology is to collect data that will be used to train robots. This argument is bolstered by Amazon’s recent unveiling of humanoid robots in its warehouses.104 Amazon has claimed that the robots are designed to work “alongside human workers.”105 But with further training, the robots will be able to retrieve bins and fulfill Amazon orders, displacing human warehouse workers altogether.106

B. Borderless Work and Inadequate Labor Protections

The advent of the internet and AI technologies makes borderless work possible and allows for the unchecked capture of worker capital worldwide. As the case of the Kenyan content moderators illustrates, the emergence of a planetary labor market107 necessitates better international legal frameworks to protect the rights of all workers. As legal scholar Tendayi Achiume argues, one core presumption of international law is the unwavering right of states to exclude noncitizens.108 This presumption that only citizens of a sovereign nation may enjoy some of its legal protections disadvantages workers in the Global South.109 And as Professor Adelle Blackett has noted, individuals who cross borders to find better employment opportunities are forced to accept “inhuman conditions” and must cope with inequalities “both within and between states.”110 These pressures necessitate a more cooperative international approach to fair labor-market access. 111

The salience of race should not be lost in this discussion. LPE scholars argue that economic relations cannot be understood without reference to “the role of atavistic status subordination, particularly racialized and gendered subordination, in the construction of capitalist social relations.”112 Geographer Mark Graham paints a racialized picture of worker exploitation, domination, and disenfranchisement:

[M]illions of jobs can now be done from almost anywhere on Earth. A mass migration of labor, but not of people. . . . Some of the impacts of this planetary labour market are being observed in the most unlikely of places. . . . [I]n a rural town in Central Africa . . . , in a place where many people still live in thatched huts and few families possess any of the technological gadgets of the contemporary world, . . . workers are helping to build some of the world’s most advanced technologies and services. In a large open-plan office with hundreds of desks and computers, workers spend eight hours a day doing highly repetitive work like matching names to photographs of minor celebrities they’ve never heard of, or identifying objects in photos of suburban America in cities that they will never go to. What these tasks have in common with the dozens of other routines performed in the room is that computers cannot yet perform them as effectively as humans.113

In theory, Graham adds, flexible geographies of production could distribute jobs across the world.114 But in practice, those geographies exert “huge downward pressure on wages and working conditions” everywhere.115 Graham argues that separating workers by large distances (as well as linguistic and cultural differences) limits the workers’ ability to associate and organize, which allows companies to exert greater power over the workers.116

Due to the rise of borderless work, the lack of adequate extraterritorial labor-law protections becomes another facet of the law and political economy of AI technologies that facilitates the capture of workers’ capital. In 2023, the U.S. Supreme Court affirmed that there is a general presumption against the extraterritoriality of American law.117 The Court articulated a two-step test for when this presumption can be overcome, looking specifically to the intent of Congress for acts to have foreign reach and also the focus of the congressional concern.118 Previously, the presumption against extraterritoriality in American law had been noted by the Court in EEOC v. Arabian American Oil Co. (Armco),which dealt with a plaintiff who was hired in the United States, transferred to Saudi Arabia, and then fired.119 He alleged that his firing was a violation of Title VII of the Civil Rights Act of 1964 (Title VII), but the Court held that Title VII did not apply abroad.120 Legal scholars have noted that the decision in Armco entrenched the presumption against the extraterritorial application of American labor and employment law.121

Recently, even evidence of criminal activity has not been found to shake the presumption against extraterritoriality. In Daramola v. Oracle America, Inc., the Ninth Circuit concluded that because the plaintiff’s employment relationship had a locus in Canada, the antiretaliation provisions of the Sarbanes-Oxley Act and Dodd-Frank Wall Street Reform Act did not apply; it did not matter that the plaintiff faced removal as a project manager and received negative performance reviews after expressing an unwillingness to participate in fraudulent activity in the workplace.122 In reaching these conclusions, the court relied on a previous decision from the D.C. Circuit, in which the plaintiff was a U.S. citizen employed overseas by foreign subsidiaries of Morgan Stanley.123 In that case, despite the fact that the plaintiff was a U.S. citizen working for the subsidiaries of a U.S. company, the court had ruled that applying the Sarbanes-Oxley Act would be impermissibly extraterritorial.124

These legal precedents show that workers in the planetary labor market do not enjoy the protection of American labor laws. As Armco makes clear, those workers are not entitled to the antidiscrimination protections of Title VII. And as Daramola makes clear, those workers are also not covered by the antiretaliation provisions under American labor law that protect workers when they report misconduct in the workplace. This leaves workers in the planetary market vulnerable, particularly with regard to the capture of their capital. Novel legal frameworks are needed for redress.

III. legal redress for workers

An LPE approach to the problem of captured capital suggests that the law should intervene to correct the lopsided power relations enjoyed by firms in the AI revolution and prevent the exacerbation of economic inequality. In the following Sections, I propose three legal frameworks that could offer workers different paths to reclaim their captured capital. These avenues for redress include (1) a corporate-governance model of data as stake capital; (2) a data-licensing regime; and (3) a guaranteed income for displaced workers. I also address some potential critiques of these proposals. These proposed legal frameworks are not necessarily mutually exclusive; two or more of these proposals may be adopted together. Finally, I argue that the International Labor Organization (ILO), as an intergovernmental agency, can play an efficacious role in promoting worker data councils under the corporate-governance model as well as enacting a guaranteed income for displaced workers.

A. Data as Stake Capital

Stakeholder capital, or “stake capital” for short, is the corporate-governance theory that those with a “stake” in a firm ought to have a say in its governance.125 Stakeholders can include employees, customers, suppliers, and creditors. The stake capital is the resources these individuals have contributed to the business, such as time or money, without actually purchasing shares of the business. Stakeholder theory suggests that these individuals should have a say in the corporate governance of the business to which they have contributed or the business that affects them.126 This theory is generally posited within the governance of a particular business and would allow directors and officers to maximize benefits to the stakeholders, rather than focus solely on profits accrued to the shareholders. Contrary to stakeholder theory, shareholder primacy had long been a feature of American corporate law.127 But some legal scholars now see this elevation of shareholder interests as an incorrect interpretation of the law.128 Under a corporate-governance model that values stakeholder interests, the business does not solely provide profits to the stakeholders directly through investment returns but would rather seek to improve its management and conditions for employees and other affected members of the community.129

If worker data is viewed as “stake capital,” workers would have a right to govern this data. This reconceptualization of worker data would bypass questions of ownership and focus on the issue of control. Treating worker data as “stake capital” would eliminate the need to parse which data is owned by the worker and which is owned by the employer. The focus would not be on an exclusive and quantifiable claim to ownership or even on exclusive control; rather, it would be on how the collective data of the firm is managed or exploited. A stake-capital theory of worker data would grant workers the right to engage in corporate-governance discussions at their firms regarding what is done with their data. Thus, in a “data-as-stake-capital” regime, one could imagine the creation of worker data councils where workers at a given firm elect representatives to bargain on their behalf regarding the collection of data and its uses. These data councils would have a separate and more particularized function compared to unions. Whereas unions would focus on labor sectors and aim to represent large swaths of workers, data councils would focus their work on one specific firm. Thus, data councils would be able to deal with the particular circumstances of any specific firm and provide more tailored approaches to data governance.

This “stake-capital” approach to worker data governance would ease the legal tension between the conception of data as bearing individualized property rights and the reality that data increasingly emanates from social relationships. As legal scholar Salomé Viljoen has argued, the primary focus of data production is “to relate people to one another based on relevant shared population features.”130 In Viljoen’s view, the transformation of information into a commodity is wrongful because it creates unjust social relations that exacerbate social inequality.131 While privacy laws in the United States primarily focus on the potential harmful effects of data on the individual,132 Viljoen contends that the paramount problem with improper data use is instead the harm to social relations at large.133 This conceptualization of “large” data, not as an accumulation of individual data but as a set of social relations, better explains how and why data collection and use produces economic value and social harm in the digital economy. A’s data can harm B, and this relationship should be properly considered.134 The “data-as-stake-capital” approach is sensitive to the communal and interconnected nature of data, as Viljoen has described.135 Through worker data councils, workers can collectively advocate for how their data will be collected and used. Such advocacy could deter data-collection efforts that are solely extractive and exploitative and promote the use of data for research that would serve workers’ interests.

Some legal scholars have critiqued “stakeholderism” as “an ineffective and indeed counterproductive approach to protecting stakeholders” and “likely to be detrimental to stakeholders and society.”136 These criticisms stem from the legal reality that corporate leaders have significant incentives not to protect stakeholders beyond what would serve shareholder value.137 Given those legal constraints, those scholars argue instead for the necessity of external interventions via new legislation, regulation, or policies to protect stakeholders.138 Taking these concerns as valid, in the following Sections, I also propose new external legal frameworks directly aimed at empowering workers.

B. A Data-Licensing Regime

The second proposed solution to the problem of captured capital is a legal framework for workers to license their data to different firms. While licensing is often considered an individual activity, consistent with the relational nature of data, this proposal envisions communal licensing schemes that would benefit workers in the same sector.

A recent collective-bargaining agreement negotiated by the actors’ union SAG-AFTRA provides a prime example of such a licensing scheme.139 On January 9, 2024, SAG-AFTRA announced an agreement with Replica Studios, an AI voice technology company.140 This agreement allows SAG-AFTRA members to be engaged by Replica under a “fair, ethical agreement to safely create and license a digital replica of their voice” and allows Replica to use the licensed voices in video-game development and other interactive-media projects.141 This agreement establishes minimum terms and conditions and provides performers with the opportunity and ability to opt out of having their licensed voice used in new works.142 Importantly, this agreement is reported to achieve “fully informed consent and fair compensation” for SAG-AFTRA members who are interested in engaging in AI technology vocal work.143 It also appears to be the hope of SAG-AFTRA that this agreement will have something of a domino effect and “pave[] the way for other companies to follow [Replica’s] lead.”144

A copy of the contract to be executed was made available to the public.145 Prominently at the top of this agreement are multiple definitions of allowable uses of licensed voices, such as development use and external use.146 The contract also establishes wage rates through January 31, 2025, for both a four-hour workday and a six-hour workday,147 and it includes a table detailing additional compensation for performers depending on the number of lines that were ultimately used in a project.148 The contract further stipulates that this additional compensation shall be paid to performers on or before the date the project becomes public.149 Finally, the agreement establishes that developers must obtain express written consent from any performer before they can use preexisting recordings of that performer’s voice.150

In even more recent news, on April 12, 2024, SAG-AFTRA announced a tentative multiyear deal with Warner Music Group, Sony Music Entertainment, Universal Music Group, and Disney Music Group.151 This agreement created the first “collective bargain guardrails [to have been negotiated] assuring singers and recording artists ethical and responsible treatment” as it pertains to AI in the music industry.152 The 2024 Sound Recordings Code was ratified on April 30, 2024, which means that the “guardrails” are now in effect.153 The agreement updates the definitions of “artist,” “singer,” and “royalty artist” to indicate that these words can only mean humans.154 The agreement also requires obtaining “clear and conspicuous” consent before an artist’s voice may be digitally replicated.155 If an artist’s voice is replicated, they are entitled to information regarding how the replication will be used as well as minimum compensation.156 Furthermore, “records labels must obtain consent on a per-project basis.”157 A major bargaining tactic that resulted in these protections for creative workers was the history-making SAG-AFTRA strike, which lasted 118 days.158

The SAG-AFTRA agreement provides a useful model for future licensing schemes for worker data because it takes care to center human workers. Its provisions focus on ensuring that human actors are not displaced altogether. The notion of clear and continuous consent is also an important part of the SAG-AFTRA licensing regime that should be emulated by other licensing regimes to ensure that workers continue to have control over their data.

C. A Guaranteed Income

Although the paramount goal remains to empower human workers in the workplace, the third proposal is a pragmatic approach that envisions a guaranteed income for workers who will eventually be displaced by AI technologies. This proposal is predicated on a set of rights and social-justice-based ideals and draws from the ILO’s principles, such as the right to just compensation and the right to fair wages, rather than a data-as-commodity framework. Specifically, the recommendation is for the ILO to request that its member countries levy a tax on companies who are planning to automate their firms. The tax proceeds would form a fund (perhaps jointly managed by the ILO and the World Bank) for a guaranteed income to be paid to displaced workers worldwide.

A corollary of this proposal can be found in the International Financial Facility for Immunization (IFFIm), which is an organization that is funded by private-sector investment.159 IFFIm ensures the availability of long-term funds for global health and immunization programs in seventy of the world’s poorest countries.160 IFFIm derives its funds from “legally binding grants payments from its sovereign sponsors,” which include the United Kingdom, France, Norway, Italy, the Netherlands, Spain, Australia, Sweden, Brazil, South Africa, and Canada.161 According to the World Bank’s website, “[w]ith the backing of these pledges, IFFIm borrows money by issuing bonds in the capital markets to fund vaccination programs in developing countries.”162

Similarly, the proposed guaranteed-income fund would depend on sovereign commitments of funds from ILO member states. Even better, the funds committed by sovereign sponsors would ultimately be financed by proceeds from an increased tax on corporate entities and thus should not burden member states. While corporate tax raises might face resistance, this proposal is not impracticable given that corporations worldwide currently pay little to no taxes on their income.163 Corporations are able to pay a miniscule portion of their income in taxes precisely because of existing laws that allow for corporate rebates and other tax incentives.164 Thus, low corporate tax payment is generally not an issue of corporations acting contrary to law or even an issue of lacking enforcement of current laws. Creating a law that mandates certain corporate taxes for this proposed fund is the first step to enforcing the payment of appropriate taxes by companies. In lieu of commodifying worker data and compensating workers for their data directly, this proposal recognizes the difficulty in quantifying the work of human beings and affirms that all humans have a right to a livelihood.

The notion of a guaranteed income has already taken root in the United States. In September 2023, two U.S. representatives reintroduced the Guaranteed Income Pilot Program Act, which would establish a nationwide pilot program as opposed to the locally funded ones currently in place.165 Representative Jan Schakowsky stated: “It is our duty to ensure that all Americans have access to fundamental rights like food and shelter . . . . This bill will help gather data about guaranteed income as an innovative way to reduce inequality and create economic security.”166 Generally, supporters of guaranteed-income programs note that there are many emergencies that can push families into homelessness and that there ought to be a new approach to helping families maintain financial security.167 Providing income to individuals facing job displacement finds precedent in governmental actions during the COVID-19 pandemic, when millions of families received cash directly from the U.S. government, leading to “overwhelmingly positive results.”168 Per the Economic Security Project, there are currently 150 guaranteed-income pilots across the United States.169

One criticism of guaranteed-income programs is that they create a culture of dependency.170 Another criticism revolves around how participants will spend money if the funds do not come with directions or limitations. Proponents refute this by noting that there is no evidence to support profligacy on the part of guaranteed-income recipients.171 Furthermore, the implementation of a guaranteed-income program is grounded in social-justice principles rather than in a commodity-based view of data. Given how worker data being collected is utilized, it will necessarily displace workers in the future.172 A justice-oriented approach dictates that workers whose data powers the AI that displaces them should be afforded a means to a livelihood.

D. Comparing the Three Proposals

The stakeholder approach is firmly grounded in corporate law, and while it would be an innovation, it is not radical. Despite the fact that the stakeholder approach might be conceptually more digestible, there remain significant issues. Even if data councils stress a collective ownership-and-control approach rather than an exclusive ownership approach, some employers as well as workers may argue for a more individualized approach. Such an approach would then require parsing and quantifying stake interests, creating snags for this proposal. Determining how much stake one specific individual has contributed versus another, and how much value may be attributable to certain data versus others, may then become an unsurmountable barrier to implementing this approach. Furthermore, there is the question of whether data governance is best left to the workers. Worker councils, which may bear no special expertise in data science or AI, may be too focused on immediate compensation to understand the far-reaching and downstream consequences of the data agreements they enter into today.

The data-licensing regime has existing models. The SAG-AFTRA agreement exemplifies what implementation could look like. However, since there is no existing union to represent all the workers who are affected by the phenomenon of “captured capital,” a licensing regime would only work if existing domestic sectoral unions are willing to take up the mantle. Additionally, it is unclear how this could work on an international scale, as there are few international unions.173 In the era of borderless work, a data-licensing regime is dependent on the advent of true international unions that can represent workers across borders. As the following Section discusses, perhaps the ILO could play an important role here.

Finally, the guaranteed-income program provides the most ambitious solution to the compensation problem posed by captured capital. While the guaranteed-income program is firmly grounded in social justice and would address many of the inequality issues discussed in this Essay, it is also highly progressive. Garnering support for this proposal would be difficult. Furthermore, although there has been some state-level experimentation with a guaranteed income, there is not yet a good national model.174 Thus, an international model might seem far-fetched until there is an organization with the platform to conduct national or even multinational pilots for guaranteed-income programs.

E. A Role for the ILO

Faced with a planetary labor market, international organizations like the ILO have an important role to play in maintaining workers’ rights. The ILO is an agency of the United Nations (UN) that works together with governments, employers, and workers to promote fair standards at work.175 With the Universal Declaration of Human Rights (UDHR) as its foundation, the ILO has developed the International Labour Standards Department (ILS).176 The ILS is tasked with the practical implementation of human-rights obligations at work.177 The ILO’s 2019 Centenary Declaration for the Future of Work confirmed social justice as its imperative.178 Four strategic objectives are identified as central to achieving social justice through the promotion of decent work: “promoting full, productive, and freely chosen employment; arranging for social protection; organizing social dialogue; and realizing fundamental principles and rights at work.”179 In 2011, the UN Human Rights Council endorsed the UN Guiding Principles on Business and Human Rights, which make clear that businesses must respect human rights. At a minimum, human rights at work encompass ideals expressed in the UDHR and fundamental rights set out by ILO.180

The ILO has a key role to play to protect workers’ interests in the AI revolution. It has already brought its attention to bear on researching the plight of Global South workers in the AI revolution.181 A 2024 ILO report examines how the rise of digital technologies, including digital labor platforms, is reshaping Kenya’s economic landscape and affecting the experiences of workers, especially women, who are engaged in this type of work.182

The ILO should continue this work in two key ways. First, similar to its advocacy for workers’ rights to unionize,183 the ILO is well-positioned to promote the development of worker data councils.184 For instance, it could develop guidance establishing how such councils should operate and what standards for the collection and use of worker data they should implement. Second, for the proposed guaranteed-income program,185 the ILO could also play a key role in collecting data and running pilots to help determine what should serve as the guaranteed minimum income in each member state as workers start to be displaced by AI automation worldwide. The ILO could coordinate its member states to impose a corporate tax on companies planning to automate their firms. The tax proceeds could then finance the fund for a guaranteed income to be paid to displaced workers.

Conclusion

AI technologies are rapidly infiltrating the business sphere. While these technologies may enable greater worldwide labor-market access, they also introduce new vulnerabilities for workers. AI technologies whet employers’ appetites for training data, and the quest for such data has enabled extractive and exploitative practices by firms. Furthermore, AI technologies may displace human workers altogether. Although a ban on the development and use of all AI technologies would be akin to King Cnut attempting to hold back the tide,186 the surge of AI technologies should not induce a techno-fatalism where we complacently accept all undesirable aspects of AI technologies, including how they enable the capture of workers’ capital. An LPE approach to corporate governance requires that the law no longer ignore the asymmetrical power relations enjoyed by firms in the AI revolution; instead, the law must address the deleterious effects of this asymmetry. The law must ensure that workers regain some measure of control over their data and can benefit from the data they create for firms. While this Essay’s proposals are not perfect solutions, they offer an LPE-guided attempt to ensure that workers do not become human scrap in the AI revolution.187

Asa Griggs Candler Professor of Law, Emory Law School. LL.M., Yale Law School. Many thanks to Professors Asli Bali, Jack Balkin, Amy Kapczynski, Daniel Markovits, and Robert Post for their helpful comments and to my research assistants, AnnaClaire Bowman and Noah Neundorfer. Many thanks also to the editors of the Yale Law Journal for their outstanding editing.