Tacoma News Tribune reporter Kristine Sherred couldn’t believe what she was seeing. A coworker wrote about a local man who takes his golden retriever with him in a motorcycle sidecar. A few weeks later she saw the article republished. This time, the writing was choppy and charmless, stripped of her colleague’s voice … and his byline. Something was off.

Sherred sent the link to her union-represented co-workers. They would eventually find out that their employer McClatchy Media had fed the article into a new AI (artificial intelligence) tool it calls the Content Scaling Agent. Powered by Anthropic’s Claude, it’s designed to rewrite and repackage articles for different lengths, formats, and audiences. Now the feature story about the sidecar dog  could have been written by anyone. It wasn’t journalism any more; it was “content.”

AI’s rapid rollout is creating similar situations for millions of workers. But like Sherred, some of those workers have a union, giving them a chance to unite and defend themselves. Sherred’s union, the NewsGuild, made AI policy a centerpiece of her union contract campaign. And across the United States, unions are responding to the threat of AI by pushing for contract language that protects workers’ autonomy and job security.

What exactly is AI?

First coined in 1956, the term AI describes computing that’s designed to mimic human intelligence. IBM defines it as any technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity, and autonomy. 

A TV remote can change the channel. But it can’t learn things about you from the way you channel surf or plan how it will help you find shows faster tomorrow. AI can. 

What we call AI lumps together several computing concepts. Machine learning is computing that crunches huge amounts of data to produce algorithms based on patterns in the data; it’s used in speech and image recognition, for example. Large language models rely on how predictable human language can be. Training on vast amounts of text, a large language model can summarize, translate, and generate apparently natural-sounding language by predicting the next most likely word in a sequence based on statistical patterns. Generative AI uses existing content as source material to create new content like text, images, videos, and music based on prompts.

Let loose in the workplace, AI can revolutionize productivity. It can also wreak havoc.

Chaos at the clinic

At Oregon Health and Science University (OHSU), AI caused a meltdown one morning in a South Waterfront surgical clinic. It’s a union member’s job to call patients, tell them when their surgery is scheduled, and make sure they prepare in the right way. But someone high up the executive chain of command decided some of that work could be handled by an AI “agent” robocaller supplied by an outside vendor, LeanTaas. One morning, all the patients at OHSU’s CHH2 building started showing up at the wrong time and place for their surgery. Luckily, these were elective non-emergency surgeries and could be rescheduled. But the incident wasted the valuable time of medical staff. And it set off alarm bells for OHSU physical therapist Angelo Bologna.

As chief steward for AFSCME Local 328, Bologna has been hearing about problematic AI projects in the workplace, introduced without the consent or even knowledge of the affected workers. For example, AI was being used in employee performance reviews, and generating inaccurate productivity metrics. Elsewhere, an AI program was brought in to help reduce a backlog in coding medical procedures for insurance billing (a backlog Bologna said was created by ill-considered employee layoffs).

Sometimes the AI projects seem to work well. Sometimes not. Bologna has come to refer to it as AI’s “blast radius.”  When an individual employee makes a mistake, the impact might be just one patient, and the employee can be disciplined or corrected. But when AI screws up, it can create problems system-wide, and there can sometimes be no one to hold accountable. 

Local 328 started making noise, demanding a say in AI’s introduction. After it appealed to OHSU’s top leadership, Bologna was added to an AI governance committee, a group of clinical informatics specialists who review and approve new uses of AI at OHSU.

Local 328 considers these AI introductions to be workplace changes over which employers are obligated to bargain with the union under Oregon’s public sector union law. Local 328’s current contract with OHSU doesn’t expire until 2028, but the changes are happening now. The union demanded OHSU bargain over them. The two sides are now halfway through a required 90-day bargaining period that runs through Oct. 14.

AI explodes at work

Workers in all sorts of workplaces are seeing AI introduced, often without notice. 

Some of the earliest impacts are in tech, where being proficient at AI tools is now a requirement for many jobs. Entry-level tech workers like engineers and designers are seeing themselves replaced with AI. According to an estimate earlier this year, AI or automation was cited as a factor in more than 185,000 tech worker layoffs in 2026. 

Often workers are confronted with AI even before they get the job. Job listing platforms like Indeed, ZipRecruiter, and LinkedIn now offer employers AI tools to screen applicants, and some companies have even begun using AI to conduct and evaluate job interviews.

At some call centers, AI systems grade how workers handle calls. Increasingly, office workers are surveilled by AI-driven software that tracks and evaluates their productivity, counting keystrokes, taking random screenshots, recording their application use, and logging the amount of time the keyboard and computer mouse is idle.

Unions fight back

To regain some control for workers, unions are pushing AI protections into their contracts.

Lisa Kresge, senior researcher at the UC Berkeley Center for Labor Research and Education, has documented over 175 examples of union contracts that try to set guardrails for the introduction of AI. Her inventory is available online at laborcenter.berkeley.edu/negotiating-tech.

The contract language is aimed at reining in several problems with AI:

• That it can replace some kinds of human labor, leading to job loss or deskilling
• That it can be used to surveil and monitor workers, often unfairly
• That workers are sometimes unwittingly training the AI that’s intended to replace them, without their knowledge or consent.

Under Oregon AFSCME’s 2025-2027 contract with the State of Oregon, state agencies are prohibited from using AI to monitor employees or using AI during hiring, promotion, worker evaluations, or discipline. The contract also requires “human-in-the-loop” oversight of any AI systems. And they’re obligated to notify the union 90 days before the implementation of any AI or automated system. 

That last item is increasingly one of the most common AI bargaining demands — just that employers give workers advance notice when they’re introducing AI in the workplace. 

At Kaiser Permanente, the 22-union 60,000-member Alliance of Health Care Unions coalition won a commitment to create a labor-management task force on technology and AI and to inform the union of any new AI technology that will be implemented for monitoring productivity or performance.

Though it’s just a start, notification is vitally important, says Oregon AFSCME rep Eddie Charlton, who represents research workers at OHSU.

“AI implementation is invisible,” Charlton told the Labor Press. “The only way to counter it is to be extremely aggressive in your investigation and requests for information from the employer…. Without guardrails, the technology can be implemented in a way where it’s irreversible.”

Human workers can strike 

Back at the Tacoma News Tribune, Sherred and her co-workers were concerned enough about AI that it featured prominently in a one-day strike May 26. Reporters at her Tacoma paper and four other Washington papers owned by McClatchy Media walked off the job carrying signs like “Keep news human.” On June 29, the union announced ratification of a new agreement that bans McClatchy from using AI to replace staff, decrease pay, or reduce working hours; from forcing reporters to use AI; from publishing deepfakes or from impersonating union members without their consent or creating fake AI journalists. Under the contract, nothing generated by AI can be published without human review, and McClatchy must disclose to readers when it uses AI.

Nationally, NewsGuild members have ratified more than 50 union contracts that  restrict the use of AI in newsrooms — holding the line for journalistic standards and protecting readers from AI-generated slop. 

In April, more than 150 journalists at the non-profit online media organization ProPublica struck in part over management refusal to agree to guardrails over AI. In July, they ratified a first union contract that includes a commitment to be transparent when AI tools are used.

But the progress is battle by battle, newspaper by newspaper. 

On July 24, reporters at the Skagit Valley Herald in Mt. Vernon, Washington, went on strike over AI guidelines (and a demand for a $20 an hour wage floor.) They returned to work July 28 without agreement, but vowed to begin an open-ended strike if there’s still no agreement by Aug. 24, the day this issue went to press.

How the Labor Press does – and does not – use AI

AI tools can sometimes be powerful aids to journalism, but our use of those tools must be guided by a commitment to accuracy and transparency and by the overriding goal of maintaining credibility and trust with our readers.

WE NEVER USE AI TO WRITE ARTICLES Every article you read here was written by a human. AI can’t attend meetings, take phone calls, decide what’s an appropriate story and what’s not. It can’t know who to call and what to ask. It can’t conduct effective interviews or develop relationships of trust with sources. Only humans can do those things — in our case humans who are paid a union wage and benefits.

WE DON’T USE GENERATIVE AI IN IMAGES We use Adobe Photoshop to edit photos and illustrations but we don’t use its generative AI tools to modify photos or create new images. Because we care about human illustrators’ ability to make a living, we don’t knowingly use AI-generated images produced by others.

WE DO ALLOW OTHERS TO SUMMARIZE OUR WORK IN SEARCH RESULTS Google and other search engines now use AI to summarize news articles. Because those summaries are the first thing to appear in the search results, they eliminate a user’s need to visit the site that published the source material. This poses a threat to us because it results in fewer visitors to our site. However we choose not to block AI web crawlers because we think it’s more important that readers get information about the labor movement than it is for us to get credit for publishing that information. We hope some users will follow the links and visit our site.

WE DO USE AI TRANSCRIPTION TOOLS. We use Otter.ai to record and transcribe interviews in real time as they occur. The way it works, the text overlays the audio recording, so the transcription can be verified for accuracy. The tool helps us during the interview refer back to something that was said, and after the interview it improves our speed and accuracy when we quote a source or use information they provided.

WE SOMETIMES EMPLOY AI FOR BACK-END EDITORIAL USES. AI platforms like ChatGPT are good at summarizing text and imitating the conventions of the most formulaic parts of journalism like headline-writing. The Labor Press sometimes uses AI to suggest headlines and meta text accompanying online posts (text that’s used to tell search engines what an article is about) but there is always human decision-making in whether and how those suggestions are implemented.

WE WELCOME READER FEEDBACK AND SUGGESTIONS ON THESE POLICIES. Email your human thoughts to [email protected]