Last week, unionized journalists at Politico won a landmark arbitration regarding AI adoption in their newsroom. In the ruling issued on November 26, an arbitrator found that Politico management violated key AI adoption safeguards that had been negotiated into the union’s contract.

Among other restrictions, the contract requires that management provide the union 60 days to bargain over any new AI technology that “materially and substantively” impacts members’ job duties. The contract also requires that if AI technology is used for “newsgathering” that it meet Politico’s “standards of journalists ethics and involve human oversight.”

In a clear-cut ruling, the arbitrator found that Politico management violated both of these terms when it rolled out two recent AI-powered editorial products.

“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,” the arbitrator wrote in his ruling, which was reviewed by Nieman Lab.

As of September, 43 contracts negotiated by units of the NewsGuild-CWA, the largest union for journalists in the U.S., included language that references AI adoption in some form. The Politico ruling marks one of the first major tests for this contract language and whether violations can be successfully challenged.

“This ruling is a clear affirmation that AI cannot be deployed as a shortcut around union rights, ethical journalism, or human judgment,” said Ariel Wittenberg, unit chair of the PEN Guild. The union represents over 250 workers at Politico and its sister publication E&E News. “This is a win for our members at Politico fighting to ensure that AI strengthens our newsroom rather than undermining it.”

In a note sent to staff this morning, Politico CEO Goli Sheikholeslami and global editor-in-chief John Harris said that they “respect the decision” of the arbitrator and will “follow through on our commitments to that process, as well as to our collective bargaining agreement.”

“At Politico, we do not plan on being left behind; we will lead the industry and deliver for our audience in this new chapter. And we will rely on all of you to do it,” they wrote in the email. “Our greatest advantage is what we already have: a newsroom full of journalists and editors who understand power, politics, and policy better than anyone else. Our original reporting, proprietary data, expert voice, and the judgment of our reporters are what makes us different.”

The decision comes months after a July hearing overseen by the arbitrator, a third party appointed by management and the union to handle the contract dispute. Nieman Lab covered the hearing at the time, which concerned two AI editorial products. One called LETO uses large language models (LLMs) to produce transcriptions of political speeches and then generate text summaries. Management used the tool to publish live summaries on the Politico.com homepage during the 2024 Democratic National Convention (DNC) and vice presidential debates.

Union members testified that they only found out about the tool after it began publishing the first night of the DNC and that they were not able to directly edit the summary text.

On this topic, the arbitrator agreed with the union’s assessment that the live summaries had produced substantial factual inaccuracies, violated Politico’s style guide, and operated without corrections or retractions. The tool, therefore, did not meet Politico’s standards.

“It must be noted that the human summaries — in terms of time, at least — substantially are handicapped by the requirements of prepublication and post-publication journalistic standards,” wrote the arbitrator.

Among several defenses, Politico leaders testified in July that AI-generated summaries were not an example of “newsgathering” and therefore weren’t subject to journalistic standards, per the terms of its union contract. Leadership testified that “newsgathering” meant “reporters going out and reporting,” including talking to sources and reviewing documents. The arbitrator rejected this line of reasoning, writing, “it is difficult to imagine a more literal example of newsgathering than to capture a live feed for purposes of summarizing and publishing.”

Politico leadership also testified that there was a disclaimer on the homepage, which read, “Live summary powered by AI.” They argued that journalistic ethics first and foremost meant transparency. On the topic of disclaimers the arbitrator wrote, “the Company has eschewed accountability for the Stylebook violations…in favor of the adoption of a significant shortcut, which was to excuse itself from the application of its journalistic standards, and instead to adopt a policy that practically amounted to caveat emptor.” 

That’s Latin for “let the buyer beware,” meaning a buyer alone is responsible for checking the quality of goods before purchase.

The arbitration also took on union complaints about another AI tool, Report Builder, which allows Politico Pro subscribers to generate several-hundred-word write-ups of policy subjects based on the site’s story archive. The arbitrator compared its outputs to newsletters written by human journalists on staff, for which reporters gather news sources before synthesizing, summarizing, and analyzing them.

While these newsletters are expected to meet journalistic standards before publication, the arbitrator found that Report Builder contains “erroneous and even absurd” outputs that are expected to be fact checked by readers after publication.

Politico leadership testified in July that Report Builder should not be subject to newsroom standards because it was built and maintained by engineers on the product team — in other words, employees “outside the newsroom.” Even so, the arbitrator found that this “human input did not include coding the AI tools to meet the Stylebook or, judging from the examples provided by the Guild, even to approximate a reasonable facsimile of compliance with it.”

Despite finding that Report Builder’s rollout was in violation of the union contract, the arbitrator stopped short of issuing a cease-and-desist order for the product until the tool is compliant with newsroom standards. He wrote that such an order would be “unduly injurious” to Politico’s business.

“It is understandable that the Company’s business side would want to ‘give the people what they want,’ but it also is understandable that journalists would want to hold the line against publication of sub-standard reporting against the pressure of consumer demand for speed and other business interests,” wrote the arbitrator.

Instead, he ordered management to enter a 60-day bargaining period over the use of the two tools. He also ordered management to negotiate a remedy with the union for its past contract violations, all with continued oversight from the arbitrator.

“Journalists, by unionizing and demanding quality for their readers, are negotiating stronger ethics, accountability and actual humans producing the news,” said Jon Schleuss, the president of NewsGuild-CWA, in a statement. “This ruling is a strong message to every media boss: AI must be implemented responsibly, transparently and through negotiation with journalists.”