New York’s LOADinG Act: Protecting Public-Sector Jobs from AI

Artificial intelligence may be one of the most important tools available to improve government performance in a generation. At a time when governments face mounting fiscal pressures, AI offers a rare opportunity to improve services without simply spending more money or hiring more people. It can help public agencies process claims faster, reduce administrative backlogs, detect fraud, improve customer service, strengthen internal analytical capacity, and deliver better outcomes at lower cost.

Yet the principal barriers to adoption are increasingly institutional and political. AI derives much of its value from its ability to automate routine tasks, reorganize workflows, and enable organizations to produce more output with fewer labor inputs. Those same characteristics can create resistance from groups whose interests may be affected by technological change.

New York’s Legislative Oversight of Automated Decision-Making in Government Act, commonly known as the LOADinG Act, provides an early and revealing example of this tension.[1] Introduced as a framework for governing the use of automated decision-making systems in state government, it paired AI accountability and transparency requirements with extensive employment protections for public-sector workers. As enacted, it prohibited state agencies from using automated systems in ways that displace employees, transfer duties performed by workers, reduce compensation, or alter certain workplace arrangements. Those prohibitions now bind most of the state’s public employers, and the accountability requirements have since been repealed.

The law’s evolution is itself revealing. Through successive rounds of revision, a statute introduced as an accountability and transparency framework for AI in state government has become, in substance, a job-protection law. The mechanics of that transformation deserve as much attention as the result, and no account has yet traced them across all four enactments—the 2024 law and the three rounds of amendments that followed.

The workforce protections were a late addition to a bill that had been originally introduced in June 2023 and that had spent a year advancing as a pure AI-governance statute. Over the 14 months that followed, those governance provisions were repeatedly enacted in public and then repealed through closed-door negotiations before they ever took effect. The workforce protections, by contrast, gradually grew, reaching from state agencies alone to school districts, public universities, counties, cities, towns, and villages—that is, to nearly every public employer in the state.

The LOADinG Act’s restrictions do more than shield current employees from layoffs or lost wages. Even if no one loses a job, or a dollar of wages, an agency may not reassign to an automated system a duty that its employees currently perform. The law, in other words, prevents AI from performing almost any task currently performed by civil servants. It also prevents almost any productivity improvements through the use of AI.

Public-sector unions and other organized interests can invoke those protections to challenge or delay the deployment of AI systems that alter existing work arrangements, even where the systems in question would improve service delivery or reduce administrative costs.[2] The predictable result is slower adoption, forgone productivity gains, and a higher-cost public sector.

This issue brief examines why AI matters for public administration, describes the origins and evolution of the LOADinG Act, and analyzes its implications for AI adoption in government. It argues that the law’s workforce protections discourage the use of AI and other productivity-enhancing technologies across New York State government. More broadly, New York’s experience illustrates how efforts to protect existing work arrangements can come at the expense of government performance.

Much of the public discussion surrounding artificial intelligence has focused on its implications for the private economy. Far less attention has been paid to its potential role in improving public administration.

Emerging evidence suggests that the productivity gains from AI may be substantial, including in areas of work that the public sector performs. Three Stanford and Massachusetts Institute of Technology researchers studied the staggered introduction of a generative AI assistant among more than 5,000 customer-support agents and found that it raised productivity by 14%, on average, with the largest gains among novice workers.[3] A randomized experiment in Science found that generative AI cut the time professionals spent on routine writing by roughly 40%, while improving its quality.[4]

Much of modern government is precisely this kind of work: processing applications, reviewing permits, answering inquiries, and monitoring compliance. Researchers at the Alan Turing Institute estimate that 84% of the U.K. central government’s complex but repetitive transactions—roughly 143 million each year—are highly automatable, and that saving just one minute per transaction would free the equivalent of 1,200 person-years of work annually.[5] A trial that gave 20,000 U.K. civil servants access to a generative AI assistant found self-reported time savings of 26 minutes per day[6] (or nearly two weeks per employee per year), and a subsequent controlled study[7] at the Department for Work and Pensions confirmed meaningful daily savings concentrated in search, summarization, and first-draft writing.

The U.S. is already moving in this direction. Federal agencies disclosed roughly 3,600 AI-use cases in 2025,[8] a nearly 70% increase from the previous year, and participation in the federal use-case inventory has nearly doubled since 2023.[9] State and local governments are beginning to follow.

The implications extend beyond administrative convenience. What is ultimately at stake is state capacity. A capable government is not necessarily a larger government. It is one that can translate public resources into effective outcomes.

Yet capturing these gains requires governments to do precisely the things that make the technology valuable in the first place: reorganize how work is done, shift routine tasks from people to systems, and deliver more with the same resources. Whether public institutions are permitted to make those changes is less a technical question than a political one. New York’s LOADinG Act shows how that question is beginning to be answered.

Although the LOADinG Act began as an AI accountability and transparency act, its accountability and transparency rules were first repealed, then restored, and then narrowed again. Its core workforce protections, by contrast, survived every revision and grew broader with each of the last two. One exception ran the other way: the most sweeping protection of all, a short-lived bar on assigning future work to machines, was removed before it took effect. Table 1 sets out the sequence traced in the sections that follow.

TABLE 1

LOADinG Act Timeline

Date	Event
June 2023	LOADinG Act introduced
June 2024	Final amendment (S.7543-B) adds workforce protections, three days before passage
Dec. 2024	Governor Hochul signs original law as Chapter 674 of the Laws of 2024
Feb. 2025	Chapter 96 substantially revises statute
June 2025	Expansion act (S.7599-C/A.8295-D) passes both houses
Dec. 2025	Governor Hochul signs expansion act as Chapter 711
Feb. 2026	Second chapter amendment (Chapter 86 of the Laws of 2026) repeals Chapter 711’s Article 5, narrows disclosure, and extends the labor protections to local employees
Outcome	Governance provisions repeatedly repealed or narrowed; workforce protec-tions retained and extended to nearly every public employer in the state

The Original Law

The LOADinG Act was introduced in the New York State Senate in June 2023 by Senator Kristen Gonzalez and in the Assembly by Assemblyman Steve Otis. Senator Gonzalez, who would also shepherd later amendments to the bill, has highlighted her mother’s “good, union job in New York City’s public schools” as part of her campaigns, and has also drawn attention to her personal background as a “tech worker” and “community organizer.”[10]

According to its sponsors, the legislation was intended to establish a framework for governing the use of automated decision-making systems by state agencies and to address concerns about accountability, transparency, and human oversight in the deployment of AI in government.[11] Specifically, the bill required agencies to disclose their use of automated decision-making systems, prohibited certain uses of those systems without meaningful human review, and mandated impact assessments before deployment.

As put forward in June 2023 and through its first amendment in January 2024, the LOADinG Act was solely a technology-governance measure with no employment provisions of any kind. The workforce protections that would become the law’s operative core appeared only in the final amendment of June 3, 2024, three days before both houses passed the bill. The change came on the Senate floor, in an amendment offered by the bill’s sponsor, Senator Gonzalez; as with the chapter amendments that followed, the negotiations behind it left little public record.

Those protections restricted the circumstances under which state agencies could deploy automated decision-making systems in ways that affected employees, their duties, or their working conditions. In particular, the law prohibited any use of an automated decision-making system that results in the “discharge, displacement or loss of position” of state employees, the “transfer of existing duties and functions currently performed by employees” to an automated system, or the “transfer of future duties and functions ordinarily performed by employees” to such a system. It further prohibited uses that would reduce employees’ hours, wages, or benefits, or infringe upon rights established through collective-bargaining agreements.[12]

The future-duties provision deserves particular attention, though it did not survive. Restrictions on displacing current employees have analogues in conventional collective bargaining. A statutory bar on assigning work that does not yet exist to automated systems does not. It reached beyond existing jobs to work that had never been performed, a category of protection without obvious precedent in U.S. public-sector labor law. The provision was repealed in the first chapter amendment, before it ever took effect, which is why the other protections are what every later revision preserved. Its initial enactment nonetheless marks the outer boundary of what organized labor sought, and briefly obtained, in statute.

The law’s provisions applied broadly across New York State government. The statute defined “state agency” broadly, reaching virtually every state department, authority, board, and office, excluding only the legislature and judiciary. Whether that definition reached the public university systems was left uncertain, an ambiguity that, as discussed later, would take two subsequent amendments to resolve.

The legislation attracted support from organized labor. The Public Employees Federation, one of New York’s largest public-sector unions, stated through its president, Wayne Spence, that the law represented “a rational and thoughtful approach to managing the state workforce given the emerging development and usage of artificial intelligence.”[13] The framing was the union’s own: a law about managing the workforce.

The New York State AFL-CIO similarly praised the legislation. In a statement following its enactment, AFL-CIO President Mario Cilento described the “groundbreaking legislation” as delivering both transparency on state agency AI use and “commonsense protections for the state workforce.”[14]

Governor Kathy Hochul signed the LOADinG Act into law on December 21, 2024, as Chapter 674 of the Laws of 2024.

The First Chapter Amendment

The original version of the LOADinG Act barely outlived its signing ceremony. The act’s core provisions carried a delayed effective date of December 21, 2025—and in February 2025, 10 months before they would have taken effect, the legislature enacted, and Governor Hochul signed, Chapter 96 of the Laws of 2025, which substantially revised the statute.[15]

The speed of the revision reflects a distinctive feature of New York lawmaking. A chapter amendment is a follow-on bill that revises a statute that the governor has just signed, on terms that the governor and legislative leaders privately agreed on before signing. Deals of this kind are struck in many capitals, but New York is unusual in its scale. Governor Hochul has amended roughly one in seven of the bills sent to her, about twice the rate of her predecessor.[16] What the public record preserves in each case is the two enacted bills, not the negotiation that reconciled them. Here the device was used to rewrite the law’s substance before any of it had taken effect.

According to the sponsor memorandum for Chapter 96 (authored, like the original act and every later amendment, by Senator Kristen Gonzalez), the amendments were intended to streamline the accountability and transparency provisions and address concerns that some provisions of the 2024 law were unnecessarily burdensome.[17]

The amendments repealed much of the law’s original governance framework, including its impact-assessment requirements and meaningful human-review provisions. In their place, the revised law adopted a narrower framework centered on disclosure obligations and a statewide inventory of AI systems maintained by the Office of Information Technology Services.

What the amendments added is as telling as what they removed: new disclosure requirements for automated tools used in employment decisions within state agencies. Agencies must post these hiring- and personnel-related tools publicly, giving employees and their unions notice of where automated systems reach into employment decisions. This transparency requirement is distinct from the workforce protections proper—the bars on displacing employees or transferring their work—which restrict what agencies may do rather than what they must disclose. The governance provisions that survived and expanded were those oriented toward the workforce rather than toward citizens or service delivery.

At the same time, the amendments carried the law’s core workforce protections into New York’s Civil Service Law nearly word for word. The prohibitions on the “discharge, displacement or loss of position” of employees and on the ”transfer of existing duties and functions currently performed by employees” were reenacted as Section 80 of that law, and the sponsor memorandum explicitly stated that the legislation would codify labor protections relating to the adoption of AI while replacing much of the original State Technology Law framework.[18]

Both prohibitions, however, attach only to the use of an “artificial intelligence system”—a term the State Technology Law defines, for AI-inventory purposes elsewhere in the statute, as a system that makes a prediction, recommendation, or decision that “directly impacts the public,” and that explicitly excludes “basic calculations, basic automation, or pre-recorded rule-based…systems.” Whether that threshold narrows the law’s reach in practice, no less than “displacement” itself, is addressed below.[19]

The State Technology Law is the statute governing the state’s information-technology functions, where the LOADinG Act was originally placed in full. Shifting the labor protections into the Civil Service Law—New York’s core public-employment statute—moved them out of a technology law and into employment law, even as the technology-law provisions themselves were scaled back.

One labor protection did not make the journey through the amendment process. The bar on transferring “future duties and functions ordinarily performed by employees,” discussed above, was deleted. The most far-reaching labor protection in the statute was thus enacted in December and excised in February, without ever having been in force. (These stages are summarized in Table 2.)

TABLE 2

The LOADinG Act Across Four Enactments

Policy Area	Original Act (Ch. 674, Dec. 2024)	First Chapter Amendment (Ch. 96, Feb. 2025)	Expansion Act (Ch. 711, Dec. 2025)	Second Chapter Amendment (Ch. 86, Feb. 2026)—Current Law
Covered employers	State agencies only (legislature and judiciary excluded); reach over the State University of New York (SUNY) and the City University of New York (CUNY) left ambiguous	State agencies only	Labor: state agencies plus school districts, Boards of Cooperative Educational Services (BOCES), SUNY, and CUNY (Education Law)   Governance: any government agency in the state	Labor: includes previous, adds counties, cities, towns, villages, district corporations, and community colleges
Impact assessments	Required*	Repealed	Restored, for all government agencies*	Repealed
Meaningful human review	Required as condition of use*	Repealed	Partially restored, within the impact-assessment framework*	Repealed
Disclosure	One-time disclosure of automated decision-making systems to the legislature*	Narrowed: annual public posting of automated employment decision-making tools, state agencies only	Broadened: annual posting of all automated decision-making tools by all government agencies, including local*	Narrowed again: employment tools only; covered local entities and universities disclose “to the extent feasible”
Public inventory	Not included	AI-system inventory added (Office of Information Technology Services)	Second inventory added for automated decision-making tools*	Second inventory repealed; AI-system inventory retained
Core labor protections (displacement, transfer of existing duties, collective bargaining)	Included*	Retained nearly verbatim; codified at Civil Service Law §80(10)	Extended to school districts, BOCES, SUNY, and CUNY (Education Law)	Retained; §80(10) extended to county, city, town, village, and district employees; university protections reenacted institution by institution
Restrictions on transfer of future duties	Included*	Repealed	Not restored	Not restored
Sunset	None	Civil Service Law §80(10) expires July 1, 2028	None in act as signed	Entire Chapter 711 framework, as amended, expires July 1, 2028

The evolution of the LOADinG Act reflected two distinct objectives: governing the use of AI in state government; and protecting public-sector employees from the workforce consequences of technological change. The chapter-amendment negotiations left little trace, so the public record cannot explain why the balance between the two shifted as it did—from an accountability and transparency law into a job-protection statute. It does show which objective survived the process.

The Expansion

In June 2025, the legislature passed an expansion act that further amended the LOADinG Act, and Governor Hochul signed it that December as Chapter 711 of the Laws of 2025. The new version of the act did two things at once. First, it restored much of the governance framework that the first chapter amendment had stripped—including impact assessments and a form of meaningful human review—in a new Article 5 of the State Technology Law, titled “Automated Decision-Making in Government Agencies,” whose requirements applied to any government agency in the state.

Second, and more consequentially, it expanded the application of the law. The original act bound state agencies alone. Chapter 711 carried the workforce protections beyond state agencies for the first time, extending them through the Education Law into the education sector—school districts, BOCES, SUNY, and CUNY. The chapter amendment that followed 14 months later carried them into general-purpose local government: counties, cities, towns, villages, and district corporations. Between them, the two enactments extend the LOADinG Act’s provisions to municipal governments and school districts, which, outside New York City, employed 452,298 full-time workers in 2022, more than 60% of them in school districts.[20]

Organized labor was actively involved in the amendment process. The Professional Staff Congress (PSC) reported that it and the New York State United Teachers (NYSUT) “worked closely with lawmakers to draft the language of the new amendments,” and it described the amended law bluntly: the statute, in the union’s characterization, will protect jobs at CUNY as public-sector employers embrace artificial intelligence. PSC president James Davis said that the union was “eager to collaborate with NYSUT and the legislators who sponsored the original LOADinG Act to extend its provisions to CUNY and SUNY.”[21] That the unions themselves cast the law as job protection is itself telling: the case that the LOADinG Act functions primarily to protect employment rests not on inference from the text but on the stated understanding of the unions that lobbied for it, celebrated its enactment, and organized to extend its reach.

The Cycle Repeats

A familiar pattern followed. In February 2026, Governor Hochul signed a second chapter amendment (Chapter 86 of the Laws of 2026), which the sponsor memorandum described as “a negotiated change to the underlying chapter.” Its first two sections repealed Article 5, on governance of the technology in its entirety, and struck the separate inventory of automated decision-making tools that Chapter 711 had created alongside it. For the second time in 14 months, every impact-assessment and human-review requirement in the law was gone—and neither had bound an agency for a single day.

The workforce protections were further extended. Chapter 711 had reached the education sector; Chapter 86 reached the tier that it had left out, extending Civil Service Law §80(10)’s bar on displacing employees and on transferring their existing duties to the employees of counties, cities, towns, villages, and district corporations.

Twice, then, New York ran the same sequence: governance provisions enacted publicly with delayed effective dates, then repealed through closed-door chapter-amendment negotiations before those dates arrived, while the core protections most valued by organized labor—against displacement and the transfer of existing work—were extended to a wider class of employers.

That trajectory was visible early. Shortly after enactment, Empire Center analyst (and now Manhattan Institute scholar) Ken Girardin suggested that the legislation was less significant as an AI-governance framework than as a set of restrictions on the state’s ability to use new technologies to reorganize work and improve efficiency.[22] Two amendment cycles later, that assessment reads less like commentary than prediction.

The LOADinG Act remains too recent to permit definitive conclusions about its effects. A search of reported court decisions, decisions of the Public Employment Relations Board (PERB)—the state agency that adjudicates public-sector labor disputes and the forum where most challenges under the law would surface—and publicly available administrative materials reveals no evidence thus far of formal disputes arising directly under the statute. That absence should be interpreted carefully: improper-practice charges, grievances, and arbitration awards are generally not public until they are decided, often a year or more after filing, so the absence of reported disputes is not the same as the absence of disputes.

Yet several developments suggest that the law’s workforce provisions are already shaping the politics of AI adoption in New York government. The evidence falls into four categories: (1) an emerging pattern of labor conflict over public-sector AI; (2) the diffusion of the LOADinG model to other governments and to the private economy; (3) the state’s uneven implementation record; and (4) the risks created by the statute’s own language.

Labor Conflict

A Nassau County dispute offers an early glimpse of the politics of AI adoption in New York. In September 2025, the Civil Service Employees Association filed an improper-practice charge with PERB alleging that the Nassau County District Attorney’s Office had deployed automated systems to perform case-input work historically handled by bargaining-unit employees, reducing members’ overtime opportunities without negotiation.[23] The union paired the legal filing with public mobilization—a “Rally Against the Bot” the following month—with its local president, Kris Kalender, insisting that AI “should be negotiated” and “definitely can’t be replacing anybody that’s actually doing the work now.”

The dispute did not arise under the LOADinG Act. When the charge was filed in September 2025, Nassau County was a local employer beyond the original statute’s state-agency scope, and the charge rested on collective-bargaining law and the parties’ contract. That is precisely what makes it instructive. Even without statutory protections, organized labor treated the reassignment of routine work to automated systems as a violation to be contested through every available channel.

Where the LOADinG Act applies, it converts that bargaining posture into a statutory entitlement. Five months after the Nassau charge was filed, Chapter 86 extended the Civil Service Law’s protections to employees of counties, cities, towns, villages, and school districts—the same class of local employer that the Nassau dispute involved. A charge of that kind filed today could proceed not only under a union contract but under statute, and the statutory protections carry an advantage that no contract does: they apply automatically, whether or not a local agreement addresses AI, and they cannot be bargained away because they are a floor set by law rather than a term subject to negotiation.

Diffusion of the Model

The LOADinG model is already spreading. Within government, as the previous section traced, Chapters 711 and 86 carried it from state agencies to school districts, public universities, counties, cities, towns, and villages.

The model is also diffusing beyond government employment, with several bills proposing similar protections against AI for private-sector workers. The proposed New York Workforce Stabilization Act, sponsored by Senators Michelle Hinchey and Kristen Gonzalez, would require private businesses to conduct AI impact assessments before deployment and would impose a surcharge of 2% of business income on corporations that displace more than a threshold number of employees through AI.[24] Separate legislation would amend New York’s WARN Act to require employers to disclose whether layoffs are attributable to automation or artificial intelligence.[25]

Whatever the fate of these individual bills, their cumulative direction is unmistakable: the principle that technological reorganization of work requires advance assessment, disclosure, and protection of existing employment is migrating from a single statute governing state agencies toward a general framework for the New York economy.

The Implementation Record

The state’s implementation record reinforces the law’s asymmetry between the governance rules that it barely enforces and the labor rules that enforce themselves.

The Office of Information Technology Services now publishes a public inventory of AI systems in use across state entities.[26] Yet an April 2025 audit by the state comptroller found that more than a year after the statewide acceptable-use policy took effect, the state had no functioning inventory of its AI systems, insufficient central guidance, and no process for monitoring compliance with the policy’s human-oversight requirements.[27] The governance obligations that survived have thus been implemented slowly and unevenly, while the workforce protections are self-executing restrictions enforceable by organized interests with every incentive to invoke them. New York built a soft regime for accountability and a hard regime for employment protection.

Emerging Risks

One might try to narrowly read the workforce provisions as channeling adoption through attrition rather than banning it. But the statute’s text reaches well beyond that reading. The prohibition on transferring duties “currently performed” by employees applies whether or not any employee loses a job, an hour, or a dollar—it restricts the reassignment of work itself, not merely its employment consequences. An attrition-based transition assumes that work can be reorganized as people leave; a bar on transferring the work forecloses exactly that.

The practical consequence of the law’s broad language on AI is uncertainty. The key terms—“displacement,”“partial displacement,”“duties and functions currently performed”—are undefined, and no court or agency has interpreted them, so managers will tend toward the broadest defensible reading: the cost of guessing wrong falls on the agency or public institution, while the benefits of a successful deployment are diffuse.

That asymmetry also extends to the statute’s threshold question of coverage. The workforce protections apply only to an “artificial intelligence system” that makes a prediction, recommendation, or decision “directly impacting the public”—the same definition used elsewhere in the State Technology Law to scope the state’s AI inventory. A generative-AI drafting assistant used by an employee who remains formally responsible for the final work product might, on a narrow reading, fall outside that definition altogether, since the system itself decides nothing. But agency counsel face the same lopsided incentives here as with “displacement”: conclude that a caseworker’s AI-drafted benefits letter, eligibility determination, or enforcement referral does not “meaningfully influence” a public outcome, and be wrong once, and the agency faces a PERB charge or grievance under a statute with no judicial gloss to fall back on. There is no comparable cost to the cautious reading. The result is that agencies will tend to treat most casework-adjacent AI use—not only tools that visibly reassign a job—as covered, for the same reason they will read “displacement” broadly: the downside of guessing wrong falls entirely on one side of the ledger.

The predictable response by agencies will be to confine AI to peripheral uses that do not touch work assignments. That response is costly because the prohibitions map almost exactly onto the technology’s value proposition. AI generates its largest administrative gains by automating routine, information-intensive work—including processing, intake,

triage, drafting, and review—i.e., by transferring duties currently performed by employees to automated systems. A regime that permits AI everywhere except where it changes who does the work permits it everywhere except where it is most valuable.

Government exists to serve citizens, not to maximize public employment. The LOADinG Act’s workforce provisions are an inversion of that principle. They treat the preservation of existing work arrangements as an end that public administration must be organized to serve, even at the cost of slower, costlier, and worse services for the public that government exists to benefit.

The benefits of restricting automation are concentrated on an organized, politically active constituency—namely, public employees—while the costs (marginally slower services, marginally higher taxes, and marginally diminished administrative capacity) are diffused across millions of citizens, none of whom has a sufficient individual stake to organize in opposition. It is a classic case of concentrated benefits and dispersed costs.

The stakes extend well beyond the state workforce that the law began with. In 20 months, the LOADinG Act’s restrictions traveled from New York’s state agencies to its school districts, public universities, counties, cities, towns, and villages. The model is now proposed for the private economy as well, and it is positioned to travel to other states.

The question that the LOADinG Act poses is therefore whether American governments will approach artificial intelligence as an instrument for serving citizens better or as a threat to be contained on behalf of those they employ. The LOADinG Act implies that it will be the latter.

Please see Endnotes in PDF

Related Tags

• # Artificial intelligence (129)
• # New York state of the United States of America (1,593)
• # New York most populous city in the United States (6,066)
• # Public Employees Federation (6)
• # AFL-CIO (119)
• # City University of New York (154)
• # Civil service (14)
• # Empire Center for Public Policy (136)
• # Government Policy (16)
• # Kathy Hochul (226)
• # Massachusetts Institute of Technology (191)
• # Nassau County (49)
• # New York Lottery (64)
• # New York State Senate (61)
• # New York State United Teachers (27)
• # Public administration (155)
• # Public sector (132)
• # State University of New York System (71)
• # United Kingdom (936)