NLRA Protections for AI-Driven Layoffs?
Introduction
Does the National Labor Relations Act (NLRA) protect workers whose jobs are replaced by artificial intelligence (AI)? Rather than wait for a binding answer to this question, unions across the country are negotiating novel contractual provisions. Most successfully, the Culinary Union of Las Vegas negotiated a collective bargaining agreement (CBA) providing increased severance pay and benefits in the event of a technology-induced layoff. The CBA additionally obligates employers to bargain in good faith over any decision to implement AI in the workplace.
The Communication Workers of America developed even more aggressive bargaining goals, adding specific privacy protections and required technology training in advance of AI implementation. This past spring, those goals were partially implemented in key bargaining wins with Microsoft. Most aggressively, the International Longshoremen’s Association (ILA) negotiated a contract that includes a prohibition on all fully automated technology.
Unions unable to secure these contractual protections have resorted to other measures. In May 2025, for instance, the Screen Actors Guild–American Federation of Television and Radio Artists (SAG-AFTRA) filed an unfair labor practice charge with the National Labor Relations Board (NLRB), arguing that Llama Productions was required to bargain in good faith before replacing human workers with AI technology.
There are myriad unresolved legal questions in this space, as the NLRB is yet to issue an opinion concerning AI-induced layoffs. Courts are sure to be slow in handing down answers, moreover, as the NLRB just regained a quorum and faces a severe backlog of cases. Adding to the delay, unions have already begun to withdraw novel cases in fear of President Trump’s appointees to the NLRB creating adverse precedent.
This Essay addresses one narrow question that is almost certain to come before the NLRB in proximate years: Must employers bargain in good faith over a decision to replace union workers with AI? The source for such a bargaining obligation could either come from the NLRA or from the parties’ CBA. This Essay highlights several CBA provisions that can alter the parties’ bargaining obligations but focuses primarily on background obligations under the NLRA. Such background obligations frame any waiver of rights in the CBA.
An employer’s motivation for implementing AI is the decisive factor for determining NLRA bargaining obligations under current law. If an employer is simply motivated by reducing labor costs, that employer must bargain with the union. Conversely, if the employer is motivated by entrepreneurship or is otherwise interested in changing business direction, there is no obligation to bargain, even if many workers will be immediately fired. The context of AI-induced job loss is likely to blend these two motivations, however, creating interpretative ambiguities for courts. Such ambiguities should be resolved through a causal test that favors requiring the parties to bargain.
I. Background: Good Faith Bargaining Under the NLRA
An employer and a union representing that employer’s employees both have an obligation under the NLRA to bargain in good faith over any “terms and conditions of employment.” This obligation applies to initial contract negotiations as well as to any modifications of the contract made during its term. Intra-term bargaining is traditionally divided into two categories: decision bargaining and effects bargaining.
Decision bargaining occurs when an employer and a union meet to bargain over the making of a decision that will impact the terms and conditions of employment. Effects bargaining occurs when the parties meet to bargain over the effects that a decision will have on the terms and conditions of employment. Whereas the parties are obligated to engage in effects bargaining almost any time there is an intra-term impact of the terms and conditions of employment, the obligation to engage in decision bargaining is caveated by critical exceptions, as explored below.
Crucially, many CBAs contain provisions that modify these background rights and obligations. Employers often bargain for “management-rights clauses.” These clauses waive an employer’s bargaining obligations with respect to certain specified actions. For instance, a management-rights clause might give management unilateral discretion to relocate workers to a new location.
Under current law, management-rights clauses are effective only when they “clearly and unmistakably” waive the union’s right to bargain over the issue at hand. In contrast, management-rights clauses that attempt to give management broad and general discretion to make unilateral changes are, under current law, largely unenforced. Thus, to waive the employer’s obligation to bargain over a decision to implement AI, the CBA would have to specify so directly. Notably, the NLRB has repeatedly flip-flopped on its standard for how specific a management rights clause must be. Under Trump 1.0, for instance, the NLRB adopted a generic “contract coverage” standard rather than the “clear and unmistakable waiver” standard.
Unions, conversely, often bargain for work-preservation clauses and, more specifically, no-subcontracting clauses. Such clauses expressly bar employers from unilaterally assigning bargaining-unit work to other parties. Work-preservation clauses do not traditionally mention AI or automation, but often bar transfer of union work to something like an “other mode of operation.” Such catchall phrases arguably encompass AI, but it is certainly preferable for unions to have AI and automation explicitly included in the work-preservation provision. Regarding no-subcontracting clauses, the NLRB has not directly considered the issue, but, as argued below, it is likely that when AI directly replaces work done by union workers, it will be considered subcontracting. Accordingly, subcontracting clauses will likely take on increased significance in bargaining as their relevance to AI becomes fleshed out by the NLRB.
If there are no provisions in the CBA that waive bargaining obligations, or if the CBA provision is not adequately specific about obligations in the AI context, then the background good-faith-bargaining jurisprudence under the NLRA will govern the issue. There is little doubt that, under the NLRA, an employer would have to bargain as to the effects of an AI-induced layoff. A layoff clearly impacts the employees’ terms and conditions of employment. However, the union would have scant bargaining power in this position, as the decision would have already been made. Decision bargaining is much more complicated. The next two Parts describe the best arguments unions and employers can make as to the duty to bargain over the decision to implement AI. As a preliminary caveat, no court has decided such a case, and as argued below, the novelty of this context could be cause for developing new law.
II. The Duty to Bargain: Implementing AI to Reduce Labor Costs
Unions’ best arguments that the NLRA mandates good faith bargaining when AI replaces union workers stem from the Supreme Court’s opinion in Fibreboard Paper Products Corp. v. N.L.R.B. (U.S. 1964). In Fibreboard, a paper manufacturing company hired an independent contractor to perform work previously done by union employees. Fibreboard argued that it had no duty to bargain over this decision, even though union members would be laid off, because hiring an independent contractor was an area of managerial discretion.
The Supreme Court was not persuaded. The Court cited its own precedent as well as NLRB decisions to paint the replacement of union workers with subcontractors as squarely affecting the terms and conditions of employment. More importantly, the Court held that the employer's subcontracting decision “did not alter the Company's basic operation,” but was rather induced by “concer[n] with the high cost of its maintenance operation.”
Crucially, the court declined to make any bright-line rules around subcontracting union work, focusing instead on the employer’s motivations for the decision at issue. To this end, the Court’s ultimate holding rested on the employer’s motivations, specifically the motivation to reduce labor costs. Notably, many early commentators and circuit courts misread this opinion by overemphasizing the importance of subcontracting within the logic of the decision.
There are two overlapping parallels between this classic NLRA case and the feared context of AI-induced job loss.
First, it can be most generically argued that replacing union workers with AI is analogous to replacing union workers with new subcontractors. This line of thinking has led some scholars to argue that “artificial intelligence is the new subcontracting,” referencing Fibreboard and concluding that employers always have a duty to bargain over the decision to implement AI.
This argument has merit, but it needs to be cabined to specific types of AI use. In particular, scholars continue to debate whether generative AI mimics human output or necessarily creates novel products. The truth is probably somewhere in between. While some applications of AI simply replace humans with an AI agent, other applications generate new output that no human could produce. Whereas the former is clearly analogous to subcontracting existing union work, the latter application lacks the key quality of continuity in employee output. Moreover, as described above, the Court made clear in Fibreboard that it was not drawing a bright-line rule regarding subcontracting. Thus, even showing that AI acts analogously to a subcontractor does not force the holding that the employer and the union must bargain over the decision to implement AI. Still, this analogy does help illuminate the connection between Fibreboard and AI, teeing up the more-precise second analogy.
Second, insofar as a given employer implements AI to reduce labor costs, without any concomitant change in the work product, the underlying motivation is parallel to that in Fibreboard. As no case has gone to trial on this question, there is limited evidence on whether cutting labor costs underlie recent decisions to implement AI. Still, given that such a showing could be made, the decision to implement AI would be analogous in both motivation and result to the subcontracting decision of Fibreboard. This would produce a strong and rigorous argument that such an employer has a duty to bargain before replacing workers with AI.
Though this is a strong argument, subsequent Supreme Court precedent has, to some extent, cabined Fibreboard by charting alternate motivations that obviate the need for the Fibreboard test. The next Part of this Essay details how employers can use that subsequent precedent to avoid bargaining over a decision to implement AI.
III. The Bargaining Exception: Using AI for Entrepreneurial Innovation
Referencing equally binding precedent, employers can argue that unions have no say in the decision to implement AI, regardless of concomitant job loss. The absolute version of this claim hinges on the successful characterization of AI implementation as a core area of entrepreneurial control.
When the Supreme Court decided First National Maintenance Corp. v. NLRB in 1981, unions and union-sympathetic scholars grew rightfully worried about the future of bargaining. First National involved an employer that suspended part of its operations, firing union workers without bargaining over the decision to close shop. The representing union tried to analogize to Fibreboard, but the Court disagreed, holding that an employer is not obligated to bargain over a decision to close part of its business. Even though workers were let go, the terms and conditions of their employment were not really at issue because it was the profitability of the business as a whole rather than the employment relationship in particular that motivated the employer’s decision.
In United Food & Commercial Workers Local 150–A v. NLRB (1993), the D.C. Circuit helpfully summarized the core distinction in First National as that between “key entrepreneurial decisions” and those “motivated by labor costs.” Along similar lines, an employer-friendly NLRB held in Otis Elevator Co. (1984) that “the critical factor to a determination whether the decision is subject to mandatory bargaining is the essence of the decision itself, i.e., whether it turns upon a change in the nature or direction of the business, or turns upon labor costs.” Scholars rightfully saw First National and its lineage as a departure from Fibreboard, but many overemphasized the change, ignoring the Court’s continuity over the decisive legal factor: employer motivation.
Employers can use First National, though, to mount a serious response to any attempted analogy to Fibreboard. Insofar as AI is implemented for the purpose of producing a new or better product—such as Unilever using AI to invent new deodorant ingredients—it affects a change in the nature or direction of business. Moreover, insofar as the adoption of AI occurs in a market context where such adoption is all but required to stay afloat, AI implementation is likely a core entrepreneurial decision. Thus, under current law, there is no facially plausible objection that a union could pose if an employer unilaterally implemented AI for entrepreneurial reasons without concern for labor costs.
IV. Looking Forward: The Gray Area of Technology
The one crucial caveat to both the union-side and employer-side arguments above is that neither Fibreboard nor First National concerned a context of technological change, much less a context like generative AI. This caveat opens a window for future courts to hold that AI-induced job loss raises questions of first impression. Creating new law would have a principled basis because technological change can be expected to create situations where employers are motivated to make unilateral changes concerning terms and conditions of employment both to cut down on labor costs and to shift business direction. A situation where dual motives are at play is not easily answered under existing Supreme Court precedent.
Both circuit and NLRB courts have faced this ambiguity of dual motives in the technology context before, but they have always sidestepped the issue, mostly because the lion's share of these opinions occurred before the employer-friendly decision in First National. In two early cases, that is, the NLRB held that the decision to automate a printing facility and to replace union workers with mechanical devices required prior good faith bargaining with the union. In both cases, the nature of the work remained the same, enabling straightforward analogy to Fibreboard. The only decision regarding automation and bargaining after First National concerned theatre projectionists whose jobs had supposedly been rendered irrelevant due to technological advances. Based on the exposing fact that the projectionists were laid off years after the new technology was implemented, the D.C. Circuit found that the employer’s motivation had strictly been to reduce labor costs. These cases give some indication of possible union-friendly outcomes, but their relevance is questionable as they were either decided before First National or involved bad-faith arguments from the employer.
It is important that courts develop new NLRA interpretation to account for the AI layoffs because, as described in the opening paragraph of this Essay, unions are worried about AI-induced job loss and are proactively negotiating against feared eventualities. Employers, too, are unsure when they must bargain, and management-side firms are offering ambiguous advice. If courts were to specifically require good faith bargaining on the front end of any AI implementation likely to result in layoffs, unions would face reduced pressure to negotiate outright bans on automated technology, as the ILA has already done. This would aid in achieving the core purpose of the NLRA: to avoid “industrial strife and unrest.” Moreover, insofar as AI can increase the economic productivity of the nation as a whole, avoiding the outright bans has clear policy upsides. However, to ensure that the benefits of the increased productivity are shared by workers and management alike, it is crucial that unions have a seat at the bargaining table when deciding how AI will be implemented.
Courts need to be careful, though, in framing bargaining obligations in the AI context because, as described above, different uses of AI reflect varying motivations. Any bright-line rule stipulating to bargain or not to bargain when implementing AI would fail to apply Supreme Court precedent. Either Fibreboard’s focus on labor costs in the subcontracting context or First National’s focus on core entrepreneurial decisions would be vitiated by any such absolute rule.
From a policy perspective, a bright-line rule on bargaining obligations either way would have related shortcomings. Requiring bargaining in all circumstances would needlessly slow down implementations of AI that are going to complement rather than replace workers. On the other side, per above, giving management unilateral power to change the terms and conditions of employment mid-contract through implementing AI and laying off workers frustrates the goal for peace in the NLRA and increases the chance that workers fail to absorb the boons of AI.
To address this variation, the NLRB could require a showing of direct causation between the introduction of AI and (projected) layoffs. Or, in the alternative, a showing of direct causal relation could establish a prima facie case for the union that bargaining is required. The employer could then rebut this prima facie case by showing that labor costs played no factor in the decision to implement AI.
Focusing on cause would be more straightforward to apply than a direct inquiry into motivations because, per recent circumstances, it is often clear that AI directly caused layoffs, while still remaining unclear what the underlying motivation for that AI implementation was. A straightforward test is critical as management-side law firms and unions alike articulate confusion regarding existing obligations for decision bargaining over AI implementation. This confusion, moreover, falls more heavily upon unions who have little direct insight into the employer’s motivations for business changes. If this anticipated confusion is not resolved with a clearer rule, the union is likely to bring litigation before the NLRB to determine decision bargaining obligations. Such a delay would impose unnecessary transaction costs on both parties and, given the information asymmetry, would offer little guarantee of a just outcome.
A causal test would effectively track employers’ motivations in implementing AI—and therefore align with current law in result—because when workers are immediately laid off, replacement is the most natural reason. Conversely, when AI implementation does not directly cause layoffs, the most-natural explanation of the implementation is that the employer is attempting to alter the work product. Such an implementation would be a core area of entrepreneurial control as it would be motivated by a change in business direction rather than by simply wanting to cut costs.
In short, NLRA precedent governing decision bargaining distinguishes decisions made to reduce labor costs from those that further a core area of entrepreneurial control. AI implementation is likely to combine these two qualities, producing ambiguous situations for courts to unpack. This Essay proposes that introducing a causal component into the legal framework to capture the essence of existing law while making decisions easier for the courts and more predictable for litigants. Regardless of the legal framework settled on by courts, it is paramount that unions be given some bargaining rights amidst workforce upheaval so that they can share in the benefits that new technology brings. Mandatory bargaining for AI-triggered layoffs allows unions to fulfill their ordained role: facilitating worker and employer cooperation to promote industry and legitimately leverage collective power.
* * *
Austin Smith is a J.D. Candidate at The University of Chicago Law School, Class of 2027.