A new analysis of the world’s most cited artificial intelligence ethics principles has identified a striking omission: collective bargaining, a cornerstone of international labor standards, has no clear equivalent in the frameworks that currently guide the development and deployment of AI systems. The finding, drawn from a comparison between five widely adopted AI ethics principles and the ethical foundations of the International Labour Organization (ILO), points to a fundamental gap between how technologists and labor advocates conceive of fairness in the workplace.
Researchers examined AI ethics principles centered on justice, beneficence, non-maleficence, autonomy and explicability, and measured them against the ILO’s framework of rights and obligations. They found meaningful convergence on three of the five principles: justice, beneficence and non-maleficence all have recognizable analogs in both systems. But on the question of collective bargaining, the comparison yields no match.
For the right of collective bargaining, however, we do not see a clear ethical equivalent in existing AI ethics frameworks.
“For the right of collective bargaining, however, we do not see a clear ethical equivalent in existing AI ethics frameworks,” the authors wrote.
The gap is not merely a matter of vocabulary. It reflects a deeper divergence in how the two traditions understand who is entitled to participate in decisions about work. AI ethics frameworks, as the paper documents, tend to center on the individual. They ask whether an algorithm treats a particular person fairly, whether it respects that person's autonomy, whether its decisions can be explained to the person affected. Labor standards, by contrast, are built on the premise that workers exercise their rights most effectively through collective organization. The individual and the group are not treated as interchangeable; the group possesses rights that cannot be reduced to the sum of individual entitlements.
What collective bargaining means for algorithmic governance
The ILO's foundational commitments, the paper notes, carry a level of international agreement and institutional backing that most AI ethics frameworks have not achieved. They are the product of decades of negotiation among governments, employers and workers' organizations, and they are supported by supervisory mechanisms that monitor compliance. An AI ethics principle, however well-intentioned, cannot claim the same legitimacy. The authors argue that the ILO's commitments “provide a level of clarity, legitimacy, and durability that most contemporary AI ethics frameworks typically do not yet possess.”
That durability matters because the deployment of AI in the workplace is not a static problem. Systems are updated, retrained and repurposed. The ethical questions they raise evolve with each iteration. A framework that rests on institutional enforcement rather than voluntary adoption is more likely to maintain its relevance as technology changes.
The convergence on justice, beneficence and non-maleficence is worth examining in its own right. All three principles have roots in moral philosophy that predate modern computing, and all three have been adapted in recent years to the specific challenges of algorithmic decision-making. Justice, in the AI context, typically means procedural fairness and non-discrimination. Beneficence refers to the obligation to promote good outcomes, and non-maleficence to the obligation to avoid harm. These map reasonably well onto labor standards that prohibit discrimination, require safe working conditions and mandate fair treatment.
Autonomy, the fourth principle, opens up a more complicated comparison. In AI ethics, autonomy usually refers to the right of individuals to make informed choices about their engagement with technology, including the right to opt out or request human review. Labor standards also value individual autonomy, but they understand it as something that collective representation enables rather than constrains. A worker who negotiates alone is at a structural disadvantage; a worker who negotiates through a union exercises autonomy more effectively. The AI ethics version of autonomy, focused on individual consent and transparency, does not capture this institutional dimension.
Explicability and the question of power
Explicability, the fifth principle, is the most novel of the five. It demands that AI systems be transparent and that their decisions be explainable in terms humans can understand. This principle is a direct response to the opacity of many machine learning models, particularly deep neural networks that produce accurate outputs through processes that are difficult to trace. Labor standards have no direct equivalent, because the question of explainability arises from the technical nature of AI rather than from the structure of the employment relationship. But the absence of a labor analog here does not mean the principle is irrelevant to workers. A system whose decisions cannot be explained cannot be effectively challenged, and a right to challenge adverse decisions is central to both individual due process and collective bargaining.
The paper does not argue that AI ethics frameworks are wrong to emphasize individual rights. Rather, it suggests that they are incomplete. An ethics framework that accounts for justice, beneficence, non-maleficence, autonomy and explicability is a good starting point, but it does not yet address the full range of ethical obligations that arise when AI is deployed in an employment context. The right to bargain collectively, the right to strike, the right to organize — these are not marginal concerns in labor law; they are its backbone. An AI ethics framework that omits them cannot claim to be a complete guide to responsible AI in the workplace.
The practical significance for responsible AI policies
The findings raise a practical question for organizations that are drafting or reviewing responsible AI policies: whether those policies account for workplace obligations such as non-discrimination, freedom of association and collective bargaining. Many corporate AI ethics statements address discrimination and fairness, but few mention unions or collective representation. The paper suggests that this is not an oversight that can be fixed by adding a line about workers' rights to an existing framework. It reflects a structural feature of how AI ethics has been developed.
The people who draft AI ethics frameworks are typically technologists, philosophers and legal scholars. The people who draft labor standards are typically trade unionists, labor lawyers and government officials. These communities do not often talk to each other, and their respective frameworks reflect their different starting assumptions. AI ethics starts from the question of what constitutes a fair algorithm; labor standards start from the question of what constitutes a fair employment relationship. The two questions are connected, but the connections are not always obvious to either side.
Bridging the gap requires more than a gesture toward worker consultation. The paper implicitly argues for a deeper integration of labor rights into the design and governance of AI systems. If collective bargaining has no ethical equivalent in existing frameworks, then either the frameworks need to be expanded or the deployment of AI in the workplace needs to be subject to labor oversight that operates independently of ethics frameworks. The two approaches are not mutually exclusive, but they point in different directions. One leads toward a richer set of ethical principles; the other leads toward stronger institutional enforcement.
How AI ethics frameworks compare with international labor standards
| AI Ethics Principle | ILO Labor Standard Equivalent | Convergence |
|---|---|---|
| Justice | Non-discrimination, fair treatment | Strong |
| Beneficence | Obligation to promote worker well-being | Strong |
| Non-maleficence | Obligation to avoid harm | Strong |
| Autonomy | Individual consent, informed choice | Partial |
| Explicability | No direct equivalent | None |
| No equivalent | Collective bargaining, freedom of association | None |
The table illustrates the asymmetry. For justice, beneficence and non-maleficence, the mapping is straightforward. For autonomy, it is partial but recognizable. For explicability, labor standards have nothing comparable because the principle itself arises from the technical characteristics of AI rather than from the employment relationship. And for collective bargaining, the gap runs in the opposite direction: a core labor right has no home in the dominant AI ethics frameworks.
The asymmetry matters because it shapes how organizations operationalize responsible AI. A company that adopts an AI ethics framework grounded in the five principles will have clear guidance on fairness, transparency and harm avoidance, but it will have no guidance on how to engage with workers collectively. That silence is not neutral. In practice, it means that decisions about AI deployment that affect large groups of workers can be made without any structured process for collective input, even when those decisions touch on issues that labor law normally reserves for negotiation.
Why the ILO framework carries greater institutional weight
The paper's observation about the durability of ILO commitments deserves careful attention. AI ethics frameworks are typically produced by technology companies, academic consortia, or multi-stakeholder initiatives. They are voluntary. They carry no enforcement mechanisms. They can be revised or abandoned at the discretion of their authors. The ILO's core conventions, by contrast, are binding on member states that ratify them, and they are subject to ongoing monitoring by committees of independent experts.
That does not mean the ILO framework is perfect or that it has solved the problem of workplace AI governance. The ILO has its own limitations, including uneven enforcement and the difficulty of adapting conventions written for 20th-century industrial workplaces to 21st-century digital labor. But the institutional infrastructure that supports ILO standards — tripartite governance, reporting requirements, complaint procedures — gives them a weight that voluntary ethics codes cannot match.
The authors' point is not that companies should abandon AI ethics frameworks and adopt ILO standards instead. It is that the two systems should be brought into dialogue. An AI ethics framework that takes labor rights seriously would not stop at principles of individual fairness. It would also ask how workers can organize, negotiate and exert countervailing power in the face of algorithmic management.
That question is becoming more urgent as the use of AI in the workplace expands. Algorithmic scheduling, automated performance monitoring, hiring platforms, productivity scoring and predictive termination tools all affect workers collectively even when they are applied individually. A delivery driver whose route is optimized by an algorithm is not the only person affected by that optimization; so are other drivers whose routes are adjusted in response, and so are the workers at the warehouse where the driver picks up packages. The collective dimension is inescapable, and a framework that addresses only individual rights will miss it.
The autonomy gap in practice
Consider the principle of autonomy. In the AI ethics literature, autonomy is often operationalized as the right to receive an explanation and the right to opt out of algorithmic decision-making. These are meaningful protections, but they assume a world in which opting out is feasible and in which the person receiving the explanation has the power to act on it. For a worker who faces algorithmic management on the job, neither assumption holds. Opting out of an algorithm may mean losing the job. Receiving an explanation of a decision does not give the worker the leverage to change the decision. Collective bargaining addresses this asymmetry by enabling workers to negotiate the terms under which algorithms are deployed, rather than simply reacting to their outputs on an individual basis.
The paper does not claim that every AI ethics framework will eventually evolve to include collective bargaining. It is more cautious than that. It documents a gap and explains why that gap matters, but it does not predict when or how the gap will be closed. The trajectory of AI ethics has so far been shaped by the concerns of the technology industry and the academic communities that support it. Labor rights have not been a central concern. Whether that changes depends on whether workers and their representatives find ways to assert their interests in the governance of AI, and on whether the broader conversation about AI ethics becomes more attentive to the institutional dimensions of fairness.
What the comparison reveals about the limits of principle-based ethics
The comparison between AI ethics principles and ILO standards also reveals a broader limitation of principle-based approaches to ethics. Principles are abstract. They can be interpreted in many ways, and they can be selectively applied. A company can announce a commitment to justice without changing its hiring practices. It can pledge to respect autonomy while deploying surveillance tools that leave workers no meaningful choice. The flexibility of principles is both a strength and a weakness. It allows organizations to adopt ethics frameworks without committing to specific institutional reforms, but it also means those frameworks can be reduced to public relations gestures.
Labor standards, precisely because they are less flexible, are harder to co-opt. A commitment to freedom of association means something concrete: workers must be allowed to form unions without retaliation. A commitment to collective bargaining means something concrete: employers must negotiate in good faith. These are not aspirational statements; they are enforceable obligations. The paper suggests that AI ethics frameworks have something to learn from the specificity and enforceability of labor standards, even if the two systems cannot be directly merged.
The convergence on justice, beneficence and non-maleficence should not be overstated. The fact that both systems recognize these principles does not mean they interpret them the same way. Justice in labor law includes procedural protections that go beyond what most AI ethics frameworks envision. Beneficence in labor law is tied to concrete obligations about wages, hours and safety, not just to a general aspiration to do good. Non-maleficence in labor law includes strict liability for workplace injuries, a standard that would be difficult to apply to algorithmic harm. The superficial alignment masks significant differences in scope and enforceability.
The most significant finding, however, is the one the paper emphasizes: the absence of any ethical equivalent for collective bargaining. That absence is not likely to be remedied by adding a new principle to existing frameworks. It requires a structural shift in how AI ethics is conceived. Ethics frameworks designed by and for individual decision-makers will not naturally produce collective rights. Those rights must be asserted by workers and their organizations, and they must be embedded in the governance structures that control AI deployment.
As AI systems take on a growing role in managing work, the gap between AI ethics and labor standards becomes harder to ignore. Companies that deploy AI in the workplace will face increasing pressure to account for collective rights, not just individual fairness. Regulators, too, are beginning to ask whether existing labor law applies to algorithmic management, and whether new rules are needed to close the gaps that the technology creates. The paper contributes to that conversation by making the terms of the comparison explicit. It shows what is missing and why the omission matters, leaving the question of how to fix it to the organizations and institutions that will have to decide where they stand.
- What did the study find about collective bargaining in AI ethics frameworks?The study found that collective bargaining has no clear equivalent in current AI ethics principles.
- Why is the absence of collective bargaining significant?It reflects a deeper divergence between individual-focused AI ethics and labor rights that rely on collective organization.
- How do AI ethics frameworks differ from labor standards?AI ethics center on individual fairness and autonomy, whereas labor standards are built on collective rights that cannot be reduced to individual entitlements.
