ProPublica's reporters struck for the right to say no to a tool

On April 8, roughly 150 people at ProPublica stopped working for 24 hours. Reporters, copy editors, videographers and business-side staff walked picket lines in New York, Washington and Chicago, carrying signs that read "Thoughts Not Bots," and asked readers not to cross a digital picket line. It was the first work stoppage in the newsroom's history.

Money was part of it. So were just cause and seniority. The demand that made the strike strange was smaller, and harder to fit on a placard. Mark Olalde, on the union's bargaining committee, told Poynter he wanted "the ability for my members to say no to using a certain AI tool and not be disciplined for it if they in good faith believe that it would introduce factual inaccuracies into their work, slow down the workflow or break ethical guidelines that the company itself has written."

Some of the most trusted investigative reporters in the country walked out, in part, for permission to decline a piece of software.

Who gets to write the rules

What set it off was a document. ProPublica had published its first AI principles the month before, and they're decent principles. They promise to disclose "meaningful" AI use to readers, and they say AI won't "replace the original reporting, analysis and judgment of our journalists." The union's complaint was about who wrote them. The Monday before the walkout, it filed an unfair-labor-practice charge with the NLRB over the "unilateral implementation of AI policy." ProPublica said it had shown the principles to the bargaining committee first, and "they offered no meaningful edits."

Management's position on putting anything about AI in the contract was candid: "It's too soon to know exactly how AI will affect our work. Rather than make promises we can't responsibly keep, we are exploring how these technologies can create more space for investigative reporting."

I believe that statement. I also think it makes the union's case for it. If nobody knows yet how the tools will behave, the person closest to the story is the only one positioned to notice when one goes wrong, and a refusal right is what lets them act on it. A principle published from the top describes an intention. A reporter who can say no without being written up is how that intention gets enforced on a Tuesday afternoon with a five o'clock deadline.

What happens when nobody can refuse

Politico has already run the experiment. It launched two AI products. One posted unedited "Live Summaries" during the Democratic National Convention and the vice-presidential debate. The other, the Capitol AI Report-Builder, sold Pro subscribers 500-word policy reports. On the day the arbitration hearing opened, the union said the Report-Builder had produced claims "that Roe v. Wade remains law" and invented lobbying positions for groups that don't exist.

The arbitrator's findings, as the union reported them, were plain enough. The summaries ran in prime homepage placements "without human editing" and stayed up uncorrected despite factual errors. The reports came "complete with headlines and citations to journalists' bylined work, without any editorial review." The tool cited the reporters' own work as its sources. Their bylines became credibility a machine spent on reports they never saw.

Politico's contract required 60 days' notice, good-faith bargaining and human oversight before tools like these went live. The arbitrator ruled in November 2025 that Politico had broken it: "If the goal is speed and the cost is accuracy and accountability, AI is the clear winner. If accuracy and accountability is the baseline, then AI, as used in these instances, cannot yet rival the hallmarks of human output, which are accuracy and reliability." In May, Politico agreed to shut down both products for good.

The terms were in a contract, and they held.

What they came back with

On July 31, the ProPublica Guild ratified its first contract: four years, with raises of 13 to 24 percent. On AI, the union's own summary lists two things. One is a mandate to keep an AI policy, with "a commitment to transparency around AI use" and "human involvement in the implementation of any AI tools." The other is an AI committee "where bargaining unit members will have a say in issues related to technology."

Neither the union's announcement nor the NewsGuild's mentions a right to refuse. The full contract text isn't public, so it may hold more than the summaries do. But in March, after the strike vote, Olalde told Nieman Lab that management had rejected the refusal protection and proposed "regular discussion" and training instead. His answer then: "I don't need contract language saying I'm allowed to have a meeting." What the union announced in July is a committee and a policy.

A policy is the weaker of the two instruments, and the people living under one say so. At The New York Times, where the Guild has been bargaining since January, management struck the union's AI proposal and offered a discussion committee; managing editor Marc Lacey called contract restrictions "imprudent." In August the paper began testing AI-written search summaries for readers, which Semafor called the newsroom's first AI text "unmediated by a Times journalist or editor." Jim Luttrell, the Guild's unit chair, told Semafor the AI policies are "meaningless," because "management can change and even violate the company's policy in any way they want." Gannett wrote the same freedom into a 2024 contract draft for its Rochester newsroom, reserving the right to change its ethics guidelines "from time to time." The following February, journalists at six other Gannett papers reached tentative deals that put a limit on AI into the contract itself.

A committee decides what the tools are allowed to do in general. Refusal decides what happens to one story, on one afternoon, when the person whose name is going on it doesn't trust what the tool produced. Readers never meet the committee. They meet the story.

Whose name is on it

In September, a Cleveland.com government reporter named Kaitlin Durbin got back from her honeymoon and found her byline on a story she hadn't written or reviewed. It came from the paper's AI-assisted desk. "Does cleveland.com think they just own my name now?" she asked on X. The paper took her name off within hours and called it a miscommunication between editors. Its editor had written in February, as Straight Arrow News quoted him, that on that desk "reporters get the final say."

This won't stay in newsrooms. Anywhere a person's name, license or signature goes on the work, the same arrangement is arriving: a tool upstream, a policy written somewhere above, and a named human at the end carrying the accountability.

Find the person in your own shop whose name goes on the work, and ask what happens to them if they refuse the tool because they think it got something wrong. If the answer is a conversation, the name still means something. If the answer is a performance review, the name has been lent to the software, and the reader has no way to know. I'd want that answer before I signed anything.

A byline is a promise that someone could have said no.