For the past several years artificial intelligence (AI) has dominated headlines — both as the next transformative technology and as a threat to the environment, civic trust, jobs and democracy itself.
The latter view is bolstered by a litany of high-profile failures. One of the more recent examples was the closure of Nota News, a site billed as a solution to news deserts that was found by Poynter and Axios to have plagiarized journalists’ work.
At the same time, organizations such as Big Local News at Stanford and UC Berkeley have used AI-assisted methods to uncover 1.5 million previously inaccessible police records, leading to more than 100 investigations.
Journalism now sits between these competing realities.
Amy Mitchell, executive director of the Center for News, Technology and Innovation, said journalism’s frustrations with technology companies are often spilling over into debates about AI.
The result, she said, is a blurring of two separate questions: whether technology companies have harmed journalism, and whether AI can help journalists serve communities in important and new ways. “That resentment carries over to some of the personal practices in a way that can weaken the pace of being able to build that technology for empowerment,” she said.
Speaking at the WAN-IFRA World News Media Congress, The New York Times Publisher A.G. Sulzberger accused AI companies of building products through “a brazen theft of intellectual property” occurring “at an unprecedented scale.” Sulzberger said the Times’ litigation against OpenAI, Microsoft and Perplexity has cost more than $20 million and cited declining referral traffic as evidence that AI summaries are replacing visits to original reporting.
While acknowledging that the Times has entered licensing agreements of its own, Sulzberger urged publishers to weigh short-term revenue against long-term value in licensing deals and to focus on pressing policymakers to defend copyright and on building their own AI products.
A review of the Tow Center for Digital Journalism database that tracks news deals and lawsuits between publishers and AI tech companies shows an uptick in the lawsuits over the past year. According to their tracking (since June 2025) there have been 13 lawsuits since June of 2024; the year before, there were eight such suits, with only four the year before.
AI at the bargaining table
That same fault line — distrust of the AI industry alongside recognition that the technology is now part of the job — runs through labor relations.
In April, roughly 150 unionized journalists and staffers at ProPublica walked out for a one-day strike after more than two years of contract talks stalled in part over AI. Days earlier, the ProPublica Guild filed an unfair labor practice charge, alleging the organization rolled out an AI policy without bargaining first.
“Since there’s no real regulation happening at the federal or state level yet,” said Jon Schleuss, president of the NewsGuild-CWA. “The regulation that is happening is happening at the bargaining table.”
Schleuss said the union now has AI-related language in 74 collective bargaining agreements, up from zero two and a half years ago. “It’s going to be in every single contract,” he said.
Schleuss pointed to the CalMatters contract as the “Gold Standard.” It commits the nonprofit to not lay off staff, cut hours or reduce pay because of AI; requires a labor-management committee to be consulted before any deal to license content as AI training data; and it mandates clear labeling when AI is used.
“I wish more would do it,” Schleuss said.
But Neil Chase, CalMatters' chief executive officer, pushed back on the idea that his organization’s contract should be the industry’s yardstick.
“It pisses me off to no end that they’re using that as an example to try to bully other newsrooms into getting the same provision in their contracts, because every organization is different,” Chase said, explaining at the time of their negotiations the nonprofit had the financial standing to make these commitments. “Our ability to say we wouldn’t need to replace any jobs with AI over the next three years is very different from what other publishers would be able to say based on their situation and their context.”
CalMatters has used AI tools for data analysis, but not writing, and has publicly posted policies governing their use. Chase traced that approach to Digital Democracy, a legislative analysis tool the nonprofit expanded after taking it over in 2019. The system identifies anomalies in voting records and other legislative data, generating story leads for reporters.
Chase also questioned labor tactics at nonprofit newsrooms. “It’s different in the nonprofit world, because you don’t have the extractive ownership that some of the bigger news organizations have,” he said.
Asked to respond to Chase’s frustration, Schleuss didn’t back down: “These protections in one newsroom can lift all newsrooms.”
Schleuss pointed to POLITICO as a recent union win. In May, an arbitrator sided with the union after the company introduced AI-generated summaries and article-generation tools without the notice and bargaining required under its contract.
Ariel Wittenberg, a public health reporter for POLITICO and unit chair of the PEN Guild, said the AI products POLITICO employed “produced factually inaccurate content or content that violated our stylebook,” without corrections.
“Workers and journalists deserve to have a say on anything that can impact our livelihood or working conditions, and AI is no different,” she said.
She noted the union isn’t reflexively opposed to the technology, saying members use AI to review large government budget documents, reformat election results and transcribe interviews. “AI that truly supplements our work can be really helpful, but it shouldn’t supplant us or our standards.”
Schleuss pointed to McClatchy’s Content Scaling Agent as an example of the kind of AI deployment the union opposes, noting that it sparked byline strikes at several papers and played a role in a recent unionization effort at Pennsylvania's Centre Daily Times.
He also rejected the idea that journalists oppose AI altogether. As a former data journalist who used AI tools at the Los Angeles Times, he drew a distinction between technology that helps reporters and tools that replace editorial judgment. “You still have to ground-truth them, and then you don’t let it write the final narrative that appears to readers,” he said. “You’ve got to have humans in control of these things.”
Schleuss echoed Mitchell’s argument that successful AI adoption requires room for experimentation. He said too many publishers are buying products from outside vendors instead of building tools with journalists. “I see a lot of teams of tech bros and other folks who aren’t working in news selling things to publishers that they’re incorporating,” he said. “The newsrooms that are doing better with AI have innovative teams.”
Building the policy with the newsroom, not for it
At the Houston Chronicle, leadership has tried to make AI adoption a conversation rather than a mandate.
“We had a newsroom survey we did, in which we told people not to hold back, and they did not pull any punches,” said Jennifer Chang, the Chronicle’s senior director of digital experimentation and innovation. The results surprised her. “The older veteran journalists didn’t have as many fears, worries or qualms. The younger ones were actually a little more worried, uncertain, scared.”
That finding sent Chang and Executive Editor Kelly Ann Scott into smaller, team-by-team conversations across the newsroom. Hearst’s editorial director for AI initiatives, Derrick Ho, said the Chronicle’s written guidelines came out of a similar process — drafted with editorial staff rather than handed down from corporate. “It’s very consensus driven,” Ho said. “It’s not a corporate overlord saying that this is how we do it.”
Scott said AI tools helped the newsroom tackle a current investigative project involving more than a million pages of documents. “There is no way we could have done this before,” she said. The time it freed up went toward what she calls “distinctly human” work, including having more time to attend community events.
That distinction between processing information and showing up in public life has long been central to community journalism. In a recent essay, Jody Chong, executive director of City Bureau, argued that while AI can help process information, it cannot replace the value of community members who show up, understand local context and create accountability through their presence. In an article titled “Why AI Will Never Replace Documenters,” she argued that civic participation depends on more than producing a record of events.
Similar questions about human judgment and accountability shape AI policy at Documented, the New York nonprofit that reports for immigrant communities in multiple languages.
Elite Truong, the outlet’s product lead, said the newsroom adapted a Poynter template built on transparency: “There’s no difference between leadership and the rest of the team in being able to suggest and experiment with new tools, as long as we check in with each other.”
Documented's policy carries an added layer because of who its journalists serve. “We work with a really vulnerable population, so we need to be very careful,” Truong said. The newsroom strips identifying information from survey responses before running them through AI tools, which are used mainly for data analysis, translation and internal project management rather than for writing or editing.
Mitchell from CNTI argues that many AI-related conflicts stem from leadership challenges rather than the technology itself. When mistakes occur, she said, the issue is often whether organizations have provided sufficient time, training and support for staff to learn the tools.
“It is a responsibility that leadership — however big or small you are, whoever’s making those decisions — that it’s time, it’s practical learning, it’s thorough learning, and it’s support,” she said.
Mitchell said one newsroom she encountered dedicated a weekly staff meeting to sharing AI experiments and lessons learned, a practice she believes helped build trust and confidence.
Mitchell said CNTI research suggests reader trust is less about the tool than accountability. “As long as you have said, ‘I put my name on this. I stand by this, and if something’s wrong with it, it is on me’ — that’s what the public cares about most,” she said.
Schleuss said he expects AI provisions to show up in every union contract going forward, and was blunt about publishers who’ve signed content-licensing deals with AI companies.
“I think every publisher who signed one of those agreements has made a huge mistake,” he said — pointing to lost compensation for journalists whose work trains the models, and to publishers negotiating “as individuals” against companies that, in his view, hold all the leverage. “We have power when we negotiate together and collectively,” he said.
Diane Sylvester is an award-winning 30-year multimedia news veteran. She works as a reporter, editor and newsroom strategist. She can be reached at diane.povcreative@gmail.com.
Comments
No comments on this item Please log in to comment by clicking here