US Government Unions Win AI Protections as Nearly 70% of Agencies Already Use the Technology
Nearly 70% of US government agencies already use generative AI, according to a survey by Ernst & Young, and public employees are demanding rules. In Oakland, a union that voted to strike won retraining instead of AI-related layoffs. In Pennsylvania, a pilot of ChatGPT Enterprise has its results checked by a human. California banned AI surveillance and automated firings in September. Vendors selling AI to government clients will have to build with a human in the loop.
AI-processed from The Decoder; edited by Hamidun News
Imagine selling an AI assistant to a city administration, only to have the municipal workers' union declare that nothing launches without a veto right and human review of the results. This is no longer hypothetical. A survey by the consulting firm Ernst & Young found that nearly 70% of US government agencies already use generative AI, meaning chatbots and assistants that draft text, process requests, and handle routine work. Because the technology has spread widely across workplaces, public-sector unions have started pushing for concrete guarantees. The Decoder reports this, citing the Guardian.
What the unions have won
The victories so far are local, but the demands look similar from one place to the next.
In Oakland, the local chapter of the union IFPTE Local 21 voted to strike in October and ultimately secured retraining programs instead of AI-related layoffs. When an algorithm takes over part of a job, the worker is not fired. Instead, they are taught to work with the new tool or moved to another position.
In Pennsylvania, Governor Josh Shapiro guaranteed the union SEIU Local 668 human oversight during a pilot of OpenAI's ChatGPT Enterprise. The pilot was not cancelled. AI went into use, but a person checks the results.
In Ohio, the Transport Workers Union won a veto over the use of autonomous vehicles. This is the toughest measure on the list: autonomous equipment cannot be deployed without workers' consent.
Laws, not just agreements
Some of the protections are written into state law. In September, California Governor Gavin Newsom banned AI programs used to surveil employees, which are known as bossware, as well as automated firings. Such software, for example, assesses how active a person is at their computer and decides what to do about it on its own. New York has protected state employees from being displaced by AI under a law in effect since late 2025.
Labor researcher Kung Feng at the University of California, Berkeley, explains the tension plainly: employers push automation to cut costs. Union leaders, including Liz Shuler of the AFL-CIO, the largest federation of unions in the US, and Lorena Gonzalez and John Samuelsen of the Transport Workers Union, are not rejecting the technology. They support AI systems where those systems help workers, but they demand mandatory human control over every use of AI in the workplace.
What developers and product managers should do
The public sector is a slow but reliable buyer. If rules such as "a human checks the result," "no automated firings," and "surveillance is prohibited" end up in laws and collective agreements, they will also appear in requirements for vendors.
Practical takeaways:
- Build in a human-in-the-loop mode: a log of AI actions, the ability to manually reverse or correct a decision, and a record of who made the final call.
- Do not sell employee monitoring features without checking local law. Such software is banned in California, and other states may follow.
- If your AI automates workflows, prepare a workforce retraining plan for the client. For unions it is a strong argument, and it helps the customer avoid a strike.
Limits of the story
It is easy to overestimate the scale. These are individual agreements in specific cities and states, not federal law. The "nearly 70%" figure comes from a single Ernst & Young survey, and the secondhand account does not show how deeply AI is embedded: in some places it is a chatbot pilot, in others part of the workflow. The timelines and verification mechanisms are not disclosed in the source.
Still, the direction is clear. AI deployment in the public sector is increasingly a matter of negotiation rather than a unilateral management decision. For teams building AI products, human oversight and transparency are moving from a line in an ethics statement to a contract clause.
- Nov 21, 2026Anthropic mathematician says Claude helped prove the Riemann hypothesis in six weeks
- Nov 21, 2026Startup Osmo Trains AI on Smells Using 10 Million Human Ratings and Generates Fragrances From Text
- Nov 21, 2026North Korea Shows First Chatbot "Source of Wisdom": Science Only, No Politics
- Nov 21, 2026OpenAI Opens ChatGPT Spaces to Plus Subscribers: Shared Documents with AI
Key Questions Answered
What the unions have won?
The victories so far are local, but the demands look similar from one place to the next. In Oakland, the local chapter of the union IFPTE Local 21 voted to strike in October and ultimately secured retraining programs instead of AI-related layoffs. When an algorithm takes over part of a job, the worker is not fired. Instead, they are taught to work with the new tool or moved to another position.
What are the key details on Laws, not just agreements?
Some of the protections are written into state law. In September, California Governor Gavin Newsom banned AI programs used to surveil employees, which are known as bossware, as well as automated firings. Such software, for example, assesses how active a person is at their computer and decides what to do about it on its own. New York has protected state employees from being displaced by AI under a law in effect since late 2025.
What developers and product managers should do?
The public sector is a slow but reliable buyer. If rules such as "a human checks the result," "no automated firings," and "surveillance is prohibited" end up in laws and collective agreements, they will also appear in requirements for vendors. Practical takeaways: Build in a human-in-the-loop mode: a log of AI actions, the ability to manually reverse or correct a decision, and a record of who made the final call. Do not sell employee monitoring features without checking local law.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.