Last summer, Politico employees wanted to send Goli Sheikholeslami a message.
Their CEO had released a “groundbreaking” AI tool to subscribers that would generate custom reports drawn from Politico articles. But employees found that the tool would hallucinate false information: In one example, the AI did not know Roe v. Wade had been overturned, news Politico itself had broken when the decision was struck down in 2022. Ariel Wittenberg, a public health reporter and unit chair of Politico’s union, told me over email that these results were generated with “legitimate queries.” When she asked the tool about former president Joe Biden’s oil policies, it described actions taken by President Donald Trump’s administration.
Employees printed out these reports and taped them around the Arlington, Virginia, office. Later, employees tested the tool further, asking it about fictitious lobbying groups, like the Basket Weavers Guild. The feature produced reports about these, too, citing Politico reporting. This May, Politico shut down the service, emphasizing that the product had always been in beta and that testing had concluded.
When I interviewed Sheikholeslami at the Politico Arlington office last week, she was sporting a fire red suit with gold jewelry and recounted the episode with a shrug. “You could go around now and there are probably flyers about something,” she said. “I will say if you want to make an LLM hallucinate, you can make it hallucinate.”
As a media CEO in 2026, Sheikholeslami must balance the needs of a media conglomerate owner—Politico is one of many US outlets owned by Axel Springer SE—that wants AI domination with an anxious and skeptical employee class that wants job security. While some media leaders have erred on the side of caution, enforcing policies that prevent or severely limit the use of AI tools, Sheikholeslami has leaned into experimentation, launching multiple public-facing AI tools and chalking up now-defunct products as data collection. When I ask about Mathias Döpfner, the German media mogul who leads Axel Springer, and his apparent AI push, Sheikholeslami doesn’t flinch. “That is his vision for the company. He is all in. So is there pressure? No,” she said. “Is there clear vision and direction? Yes.”
She doesn’t disagree with it. “If you don’t participate, you don’t learn,” she says, hitting her hand on the conference table to underscore her point. “I would say the lesson I have learned in my career is you have to lean into change—because you have to understand, ‘How do I actually take advantage of this new technology to help me do what I want to do better?’”
Sheikholeslami’s strategy is that of a seasoned pragmatist. Ever since earning her MBA from the University of Virginia in the early ’90s, she’s worked as a media executive. She’s seen the internet slash paper sales, and social media gobble up content, then traffic. To many in the journalism industry, the fight against AI feels similarly futile, like a child landing punches against a UFC fighter. Most journalism companies can’t financially compete with well-capitalized AI companies, many of which have shown their ability to gut media’s distribution channels. AI summaries are siphoning traffic from home pages, and more and more people are getting their news from ChatGPT. When I offered that media companies just keep learning that tech companies don’t care about them, she agreed. “Categorically, as much as they say they care, they absolutely do not.”
