What happened

OFF Radio Krakow is a small offshoot of Radio Krakow, the public broadcaster serving the Krakow region of southern Poland. It is funded, like the rest of Polish public media, by the taxpayer. In October 2024 it relaunched with a gimmick that managed to combine two separate bad ideas into one news cycle: it replaced human presenters with AI-generated personalities, and then it used generative tools to put words in the mouth of a dead Nobel laureate.

The station introduced three synthetic hosts presented as young people - characters with names, ages, and personas, pitched as a way to reach a "Gen Z" audience the station said it was otherwise failing to attract. They were not real people. They were voices and identities produced by software, reading copy and conducting "conversations" that no human presenter had spoken.

This followed the station letting go of roughly a dozen journalists in August 2024, weeks before the AI relaunch. Management's official explanation was that listenership had fallen to almost nothing and the dismissals were a response to that, not a swap of humans for machines. The timing made that distinction hard to sell. You do not lay off your on-air staff and then unveil a roster of AI presenters and expect anyone to treat the two events as unrelated.

The Szymborska "interview"

The detail that turned a labor-and-novelty story into a national scandal was the station's decision to broadcast an AI-generated "interview" with Wislawa Szymborska.

Szymborska is not a minor figure. She won the 1996 Nobel Prize in Literature, is one of the most celebrated poets in Polish history, and died in 2012. There was, obviously, no interview. The station used generative tools to manufacture a conversation with a woman who has been dead for more than a decade and who never consented to any of it because she could not. Reporting on the episode noted the station had obtained some form of permission from her foundation, but that does not change what the segment actually was: a fabricated exchange attributed to a real, named, revered, deceased person.

This is a useful place to be precise about why this lands differently from, say, a chatbot giving a wrong answer. A hallucinated fact is an error. A synthetic interview with a named dead Nobel laureate, aired by a public broadcaster as content, is a deliberate production decision. Someone scoped it, someone generated it, and someone scheduled it. The "AI did it" framing does a lot of unearned work in stories like this. The AI generated the audio. People decided to air it.

The backlash

Mateusz Demski, a film critic and one of the journalists who had hosted programming at the station before the cuts, published an open letter opposing the replacement of staff with artificial intelligence. He framed it not as a personal grievance but as a precedent: if a public broadcaster can quietly substitute synthetic presenters for working journalists, the same logic travels to every creative field that produces words, voices, and ideas for a living.

The letter found an audience fast. An accompanying petition gathered more than 15,000 signatures within roughly a day, and the number kept climbing over the following days as the story spread. Demski reported being contacted by large numbers of people, many of them young - the exact demographic the AI hosts were supposedly designed to win over - who objected to the experiment.

The reaction reached the national government. Krzysztof Gawkowski, Poland's deputy prime minister and digital affairs minister, weighed in publicly, saying that "certain boundaries are being crossed more and more" and using the episode to argue for AI regulation. When a deputy prime minister is citing your radio relaunch as Exhibit A for why the country needs new laws, the experiment has not gone to plan.

The station's own advisory board was reported to be appalled by the initiative. This was not a case of external critics versus a unified newsroom. The objection came from inside the building as well as outside it.

How it ended

The AI-presenter project had been announced as a roughly three-month trial. It lasted about a week.

Marcin Pulit, the editor-in-chief, halted the experiment after the wave of criticism. His public framing was that the station had gathered enough observations, opinions, and conclusions in that week to conclude there was little point in continuing - which is a measured way of saying the backlash made the project untenable. He also pointed to the episode as evidence of a regulatory gap and of genuine public anxiety about AI displacing human workers, which is true, even if the station was the one that generated the anxiety in this instance.

The names and exact ages reported for the three synthetic hosts vary slightly across coverage, and the precise running total on the petition climbed over the days that followed the initial 15,000-plus surge. The load-bearing facts are not in dispute: a publicly funded broadcaster cut human journalists, launched AI presenters, aired a fabricated interview with a dead Nobel laureate, and reversed course within about a week under sustained public and political pressure.

Why it matters

There are two failures stacked on top of each other here, and they are worth separating because they teach different lessons.

The first is the labor-and-trust failure. A public broadcaster exists on a slightly different footing than a commercial one. It is funded by the public and carries an implicit promise that there are human journalists doing journalism behind the output. Replacing them with synthetic hosts, whatever the listenership numbers, breaks that promise in a way audiences notice and resent. The "we were failing to reach young people" justification was undercut immediately by young people showing up to sign a petition against the idea. The target demographic did not want to be addressed by a machine pretending to be one of them.

The second is the fabrication failure, and it is the more serious one. Using generative tools to produce a fake interview with a named, dead, real person is not an editorial experiment - it is the manufacture of something that did not happen and the broadcast of it as content. Even with a foundation's blessing, the segment trades on the authority of a real Nobel laureate's identity to lend weight to words she never said. That is exactly the capability that makes generative audio dangerous in a news context: it produces fluent, plausible, attributable-sounding material with no underlying reality. A newsroom's entire value proposition is that what it airs is real, or clearly labeled when it is not. OFF Radio Krakow inverted that.

The encouraging part of the story is the immune response. The reversal did not come from a regulator, a lawsuit, or a slow internal review. It came from a former employee writing an open letter, tens of thousands of people signing a petition within a day, the station's own advisory board objecting, and a government minister naming the line that had been crossed. The experiment was pitched to last three months and was killed in about a week by people who simply did not accept the premise.

The broader pattern is by now familiar from newsroom after newsroom: an outlet adopts generative tools to cut costs or chase an audience, obscures or underplays what it is doing, gets caught, and retreats. The technology is not the part that fails. The decision to deploy it in place of - and in the voice of - real people is.