US Public-Sector Unions Win Curbs on AI Surveillance and Firing

US public-sector unions are increasingly winning protections against AI surveillance and against job losses caused by automation. The German outlet The Decoder reported this on 13 November 2026, citing the Guardian. The issue is becoming more urgent: according to a survey by Ernst & Young, nearly 70 percent of US government agencies already use generative AI, meaning systems that produce text, images or code on their own.

One term up front: "bossware" refers to monitoring software that employers use to watch or evaluate staff. Most of the cases below concern such software and automated dismissals. Which of the measures are legally binding and which are only commitments from the employer side differs from case to case.

What have the unions achieved so far?

After a strike vote in October, the union IFPTE Local 21 in Oakland secured retraining instead of AI-driven layoffs, according to The Decoder. In Pennsylvania, Governor Josh Shapiro guaranteed the union SEIU Local 668 human oversight during the testing of OpenAI's ChatGPT Enterprise. The transport union TWU won a veto right over autonomous vehicles in Ohio, The Decoder says.

Lawmakers are acting as well. According to The Decoder, California Governor Gavin Newsom signed laws against AI surveillance and automated dismissals in September, and New York has legally protected state employees from displacement by AI since the end of 2025. Labor researcher Kung Feng of the University of California, Berkeley, says employers want to cut costs through automation, according to The Decoder. Union leaders such as Liz Shuler (AFL-CIO), Lorena Gonzalez and John Samuelsen (TWU) welcome helpful safety systems but demand human control over every AI application in the workplace, according to the Guardian.

What does the California law actually do?

The headline "bans automated firing" is a simplification. According to KQED, Newsom signed several laws on 30 September 2026 to protect California workers. The central one, SB 947, does not prohibit all automation. It requires that dismissals or disciplinary decisions not rest solely on an automated decision system. A human must corroborate the decision, and workers must be told when such a system was used. The law takes effect on 1 July 2027. According to KQED, Newsom had vetoed an earlier version the year before and now signed a narrower one. The bill's author, state senator Jerry McNerney, argued in essence, per KQED, that nobody wants to be fired by a machine.

Other laws from the same package, signed on 30 September according to KQED:

• SB 951 (Reyes): employers must give notice when AI causes a mass layoff, relocation or termination. According to the Transparency Coalition, the notice period is 60 days when 25 percent or more of a workforce is displaced. The state agency EDD is also to study AI's effect on hiring and employment.
• AB 1331 (Elhawary): bans surveillance tools in workplace bathrooms.
• AB 1883 (Bryan): restricts AI-powered workplace surveillance, including tools that infer workers' emotional states or collect neural data.
• AB 1979 (Bonta): AI may not replace the clinical judgment of licensed health care professionals and remains advisory.

Not everything became law. According to KQED, Newsom vetoed AB 2656 the same day. It would have required notice to public employees when AI performs work within the scope of their jobs, which is the rule most closely tied to the public sector. AB 2575 also failed; it would have protected nurses from retaliation when they override AI recommendations they judge unsafe. According to KQED, rideshare drivers are not covered by SB 947 because they are classified as independent contractors under Proposition 22.

What does an agreement look like in practice?

The Pennsylvania example is well documented. According to the organization Partnership on AI, Governor Josh Shapiro published a letter in March 2025 with eight AI commitments and at the same time negotiated a binding side letter with SEIU Local 668. According to Partnership on AI, the union represents nearly 10,000 state employees. The agreement sets up a joint labor-management group on generative AI and contains further protections for workers.

From conversations with both sides, Partnership on AI, whose case study dates from 9 April 2026, draws several lessons. Defining AI as a tool rather than a replacement lays a foundation for addressing displacement risks. Workers should be involved as experts. The union advises engaging early, even without an invitation, and seeking broad agreements that can be extended to other unions through "patterning" or "impact bargaining" (negotiating over the consequences of a decision).

At the federal level, the picture remains thin. According to KQED, Congress has not passed comprehensive federal AI regulation, and President Donald Trump announced that top tech leaders signed a voluntary accord. KQED also recalls that DOGE, the short-lived initiative run by Elon Musk, used algorithmic tools at the start of the second term to identify and terminate employees. Probationary workers were fired via identical template letters, without individual review. This background helps explain why unions are betting on contracts and state laws.

What does this change for companies in the US and abroad?

The US rules are directly binding only for employers in the states concerned, for example subsidiaries of foreign firms in California, where SB 947 applies from July 2027. They still serve as a guide: a company that uses AI in hiring, firing or performance monitoring can already decide that a human reviews every dismissal and warning, that those affected are informed, and that employee representatives are involved early. This is the editorial team's assessment, not legal advice.

It remains open how strictly these commitments will be enforced. SB 947 only takes effect in July 2027, and The Decoder gives no contract details for Oakland or Ohio. The information on Oakland, Ohio and New York comes solely from The Decoder's report, which we could not verify independently.