Public-Sector Unions Are Writing the AI Rulebook That Legislatures Won't

With nearly 70% of government agencies deploying generative AI, public-sector unions are using contract negotiations to impose the guardrails policymakers have failed to deliver.

Nearly 70% of state and local government agencies have adopted or are implementing generative AI, according to an Ernst & Young survey conducted from July to August 2026. The technology is no longer a hypothetical threat to public-sector workers — it is present in unemployment offices, transit systems, schools, and hospitals. What's emerging is not a technology story but a governance story: who gets to set the terms under which AI enters the workplace.

The playbook is crystallizing fast. Oakland's International Federation of Professional and Technical Engineers Local 21 authorized a strike on 22 October with 94.1% in favor, forcing the city into a three-year tentative agreement requiring retraining or reassignment — not firing — of workers displaced by AI. The Transport Workers Union secured veto power over autonomous vehicle deployment at the Central Ohio Transit Authority in 2024. Pennsylvania's SEIU Local 668 pushed Governor Josh Shapiro into a written policy guaranteeing employee input and supervision over state AI use, with a commitment that the tech won't be used to discipline workers. The Communications Workers of America won AI protections for California workers in October 2026. The legislative track is catching up, but unions got there first. California's No Robo Bosses bill, signed by Gavin Newsom on 30 September, bars employers from relying solely on automated systems to discipline or fire workers. A companion bill prohibits AI workplace surveillance tools that monitor employees' emotional states. New York passed a law effective December 2025 protecting state agency workers from AI displacement. But these are the exceptions. As AFL-CIO president Liz Shuler put it: "We're using our contract negotiations to institute guardrails that policymakers won't." The employer-side pressure is real and growing. UC Berkeley's Labor Center documented the spread of what researchers call "bossware" — AI tools that rank workers for hiring and firing, track location and activity, scan biometrics, monitor productivity, use cameras and wearable sensors, and deploy analytics to predict emotions. Amazon's algorithmic management of warehouse workers, which pushed employees to operate at dangerous speeds, is the template now migrating into public-sector workplaces. The surveillance infrastructure is already built; the question is who controls it. Transport is the sharpest edge of the fight. TWU president John Samuelsen describes automation as "in every single element of passenger transport" — from autonomous vehicles that Florida and New York workers are trying to block, to airlines experimenting with drone inspections of aircraft rather than human jet mechanics. The union draws a clear distinction: collision avoidance and timing systems are welcome; removing human oversight is not. The pilot analogy is instructive — computers have flown planes for years, but two pilots remain in every cockpit. The structural pattern here matters more than any individual contract. Unions are establishing a precedent that AI deployment requires negotiation, not just notification. Advance notice before rollout, disclosure of technology capabilities, and the right to bargain over implementation are becoming standard demands. This is contract-level governance filling a vacuum that federal and most state legislatures have left open. Whether these protections hold depends on whether the labor movement can maintain bargaining power as AI capabilities accelerate. The honest tension is this: some of these protections will prevent genuine efficiency gains alongside preventing genuine harms. The Oakland spatial data administrator's observation — "sometimes with a fancy new toy, it actually costs more, it's more difficult to implement and the results are not as good" — captures the current reality for many government AI deployments. But that reality will change as the technology improves, and the contracts being signed now will shape whether the adjustment is managed or catastrophic.