How New York Constrains Public Sector AI
David Tan

That could make cities fall behind

While we are far from understanding the long-term implications of AI on work, changing technology has been impacting how people do their jobs for generations. Concerns about job stability, efficiency and the humanity of workers are not new in the public sector; in 1911 and 1912, machinists and other workers at government arsenals and navy yards protested being timed with stopwatches, and Congress later barred stopwatch time studies in those public facilities; in 1962 the Civil Service Employees Association (CSEA), concerned about “Electronic Data Processing programs,” asked New York Gov. Nelson Rockefeller to agree to protections for displaced employees, including job transfers, salary protection, first claim on new jobs and training programs. Today, computers, stopwatches, GPS and more are used every day to deliver government services, and the protections CSEA asked for are commonplace alongside the introduction of new technology. But New York state has drawn a line in the sand when it comes to artificial intelligence, restricting agencies at every level of government from handing work employees do today to AI systems if there is a chance it could “meaningfully influence” outcomes for the public. While that sounds narrow, key terms are undefined and the law could cover much of what agencies would want AI to help with. This has major consequences for New York City, and it should not stand. 

The Legislative Oversight of Automated Decision-Making in Government (LOADinG) Act started life as a tech governance proposal for state government in June 2023, but by February 2026, after a series of amendments, it had transformed into a wide-ranging labor protection statute applicable to public-sector workers at nearly every level of government. The law bars any use of AI that results in “transfer of existing duties and functions currently performed by employees” to an artificial intelligence system, along with any displacement. These provisions are set to expire in July 2028 unless renewed. 

The law relies on a broad definition of “artificial intelligence system” that covers generative AI and other tools that may meaningfully influence outcomes for the public. An earlier version of the law, signed in 2024 but later amended before it took effect, went further and covered any software that “support[s]…human decision-making.” Key terms like “existing duties and functions,” the crux of the law’s ability to impact how agencies work, are never defined, and it isn’t clear how the law applies to work contracted out to vendors, a key way the government delivers public services. 

As a result, these broad and vague prohibitions have the potential to severely curtail New York City’s ability to explore productivity gains or improved service delivery with AI. Agencies are already stretched thin, but duties don’t disappear when positions sit empty. Work gets reassigned to staff who are already overloaded, and under a strict interpretation of the law, that work still counts as an “existing duty.” Handing any part of it to AI could be challenged, regardless of backlogs. The Human Resources Administration only processed 65.8% of cash assistance applications on time in FY 2026, but may hesitate to use automated tools to flag illegible documents in benefits applications. Staff manually enter data from scanned applications, invoices or inspection photos. AI can save staff time and reduce backlogs, but in New York, will it be an option?  

New York state isn’t alone in looking to protect government workers from displacement that may come with changing technology, but the LOADinG Act is especially strict — in a way that is likely to constrain innovation. 

In Pennsylvania, Gov. Josh Shapiro outlined eight AI protections and provisions for state employees, while simultaneously negotiating with Service Employees International Union (SEIU) Local 668, which represents nearly 10,000 Commonwealth employees, to define additional mechanisms for ongoing labor-management collaboration on AI governance that can adapt as the technology evolves. Pennsylvania’s commitments acknowledge the potential of AI to enhance productivity and job satisfaction, while also explicitly stating that there is no desire “of substituting Commonwealth employees with AI,” instead committing to maintain a “human-in-the-loop” process. The state explicitly defines AI as a tool, not a replacement for staff, stating that “generative AI tools are never personified as ‘AI workers’ or ‘AI employees’ …no matter how sophisticated it becomes in exhibiting humanlike qualities.” The agreement also commits to including union-covered employees in future pilot programs, and finding ways for employees to use time saved by using AI for certain tasks in other ways that improve “customer service, customer outcomes, and…operations as a whole.” While the New York law restricts AI based on work employees do today, Pennsylvania embraces AI as a tool that can help save employees time and improve how services are delivered. 

In California, Gov. Gavin Newsom signed a package of bills, many backed by labor unions, covering both public- and private-sector workplaces, ensuring real people review automated employment actions and creating a regulatory governance, transparency and data privacy framework. California has also required that some public roles, like Cal State faculty and K-12 school staff, be filled by people, while explicitly leaving room for staff to use AI as a tool. It doesn’t bar public agencies from using AI for work employees do now. In Ohio, the president of the Ohio Civil Service Employees Association (OCSEA), a public-sector union representing 27,000 state employees, “advised [members] the best way to respond to AI advancements is to get training,” and its 2024-2027 negotiated contract ensures the union has a seat at the table with regard to how AI is used in state agencies. Nationally, the American Federation of State, County and Municipal Employees (AFSCME) passed a resolution on Artificial Intelligence in the Workplace at its August 2026 convention, stating that it will “advocate that workers have access to opportunities that AI creates through education and training, and where workers’ jobs are impacted by the technology, they receive protections, support and retraining as needed.” In all of these cases, unions have called out the need to support employees as work changes, and pushed for provisions that protect employment, rather than drawing a hard line at how AI can and cannot be used. 

Only New York, it seems, is explicitly protecting the duties an employee is asked to perform, not the employee themselves. The law is vague, and there have not yet been any court or Public Employment Relations Board (PERB) rulings to define scope, such as what is included in “duties currently performed” or whether agencies can let AI absorb work as employees retire. Other states are focused on building training programs and understanding when and how AI can best be used to deliver better public services in cooperation with organized labor. In New York, potential use cases have a high likelihood of getting challenged in court or through PERB. The likely result? Agencies avoid using AI, and public employees lose a new tool that has the potential to positively impact service delivery, leaving New Yorkers behind. 

Cities and states that are more open to AI are seeing the potential to improve government productivity and save taxpayers money. In Austin, Texas, city staff built AI agents to map out the component parts of the city’s permitting system, a process that could take years to do manually, or tens or hundreds of thousands of dollars to do with external consultants. Instead, the team created four prototype AI agents themselves to map the existing code and understand interdependencies that are vital for a smooth transition to a new system. Building the agents took weeks and cost only $1,000. Boston is using AI along with open data and other modern software approaches to proactively manage city curb space to make streets safer, more accessible and easier to navigate. Neither project is clearly covered by the LOADinG Act, but permitting and curb space are high-profile, high-stakes public issues in NYC. With the law’s key terms undefined, an agency lawyer could reasonably advise against starting either one. 

Only New York, it seems, is explicitly protecting the duties an employee is asked to perform, not the employee themselves.

In Arizona, the Department of Economic Security is using a human-led, AI-assisted workflow process to produce translations for SNAP and other key public benefit programs in 15 languages that meet the state’s threshold for translation. Even with its staff of translators, the demand far outpaces the team’s capacity. While New York uses AI in translation today, a strict interpretation of the law could very well argue that only human translators should do that work, regardless of existing capacity. 

In Pennsylvania, the Department of Human Services is using new AI tools to help caseworkers catch common errors in benefits applications and renewals. Documents are scanned for correctness and legibility upon upload, cutting down on time spent by caseworkers deciphering poor scans, or back-and-forth with applicants about the need to resubmit. In a two-month pilot with a portion of applicants, the tools decreased instances of illegible or incorrect documents by 80% and were estimated to save staff more than 700 hours of work. 

Under New York state’s current laws, depending on how unions and managers choose to interpret statutes, many of these use cases may not be allowable. That is especially inconvenient for Mayor Zohran Mamdani, who has made it clear that both efficiency and effectiveness of government services are a top priority. If the newest tool in the toolbox is heavily restricted, how quickly can New York City make real strides? While a search of public records by the Manhattan Institute’s Sean Speer found that there is no current evidence of formal disputes under the LOADinG Act, it has been less than a year since the statute was extended to local government, and typically grievances or other actions are not made public until finalized. 

The regulations on AI are out of step with history. For years prior to the LOADinG Act, New York City had been regulating automated decision systems (ADS); the New York City Automated Decision Systems Task Force was established by Local Law 49 of 2018 in order to provide recommendations on how City agencies used ADS, reviewing how algorithms and other systems were being implemented. New York City’s approach has been more aligned with the national approach to AI — cautious about the impact on the workforce and on New Yorkers while recognizing the potential benefits. Mayor Bill de Blasio established the Algorithms Management and Policy Officer position through an executive order in 2019 to guide the City in developing responsible use guidelines that balance ethical concerns with innovative uses to deliver the greatest benefit for New Yorkers. In October 2021, the City’s chief technology officer, John Paul Farmer, released the first New York City Artificial Intelligence Strategy, which recognized that “AI will change the world” and “touch virtually every area of life in the years ahead,” while calling for New York City to be part of shaping this future by better understanding the technology and grounding strategy in the framework of digital rights including “privacy, accountability, trust, transparency, and fairness and non-discrimination.” Building on the AI Strategy, the Office of Technology and Innovation (OTI) released the New York City Artificial Intelligence Action Plan in October 2023 that continued to emphasize the need for internal and external governance frameworks, and the need for continued knowledge building and learning by doing. In 2024, OTI also established a City steering committee and an external advisory network, and released guidance on generative AI use. Today, the LOADinG Act’s parameters take precedence over established City policy. 

Interestingly, the LOADinG Act specifically speaks to work currently performed by employees today; it remains to be seen how “duties and functions” will be interpreted by unions and courts across the state. The fact that the law does not touch on how AI or automated decision systems are used by government contractors, which perform billions of dollars’ worth of public-sector functions throughout the city and state, could potentially push work managed in-house to be outsourced to vendors to gain efficiencies. 

It is clear that government has a long way to go in learning from failed government technology projects and effectively using technology to deliver better services. Improving service delivery isn’t just about how technology is used, but about taking a holistic look at government operations, staffing, spending and how the work gets done to make the most of the capacity — the government workers — that keep the city running. This month’s election results made it clear that there are real questions about AI that we still need to answer, but the protections people are asking for are about the environment, their communities and the workers themselves, not about locking in the work government employees do today. Without rethinking how the LOADinG Act impacts City service delivery, the chances of New York City falling behind will only grow as the technology evolves.


Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to Vital City.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.