AI-written Holocaust speech? FAZ pulls op-ed by Thuringia's state premier

What happened

The Frankfurter Allgemeine Zeitung (FAZ) has removed an op-ed by Mario Voigt (CDU), the state premier of Thuringia, from its website and its archive. The paper's reasoning: the State Chancellery in Erfurt couldn't convincingly assure it that the text hadn't been written by an AI. The piece argued for banning smartphones for young people. According to heise online, the FAZ was also responding to an investigation by the platform FragDenStaat, which examined many of Voigt's speeches and op-eds for signs of AI. The findings are sensitive because they include a speech commemorating the victims of the Holocaust.

How it started: an op-ed about German music on the radio

The affair started with an op-ed in the daily Die Welt that Voigt co-wrote with Sven Schulze (CDU), the state premier of Saxony-Anhalt. It called for more German-language music on the radio, which is a fairly harmless topic. The piece only drew attention when a government spokesperson admitted, when asked, that AI tools had been used to write it.

FragDenStaat then looked at other texts Voigt had used to weigh in on public debates or to commemorate the crimes of National Socialism. FragDenStaat is a nonprofit platform best known for freedom of information requests and investigative reporting.

What the analysis found

For its investigation, FragDenStaat used Pangram. It's an AI detector: software that looks at typical language patterns to estimate whether a text was written by a person or by a language model like ChatGPT. The team analyzed eleven speeches and four op-eds. Here's what FragDenStaat found:

The most explosive finding concerns a speech commemorating the mass murder in the concentration camps. According to the analysis, it was "100 percent AI-generated." The analysis points to repeated negations and vague, hard-to-pin-down imagery. This pattern ("not X, but Y") is widely seen as a hallmark of text from large language models. In the speech, Voigt said, for example:

"Auschwitz was not the work of an imaginary monster. It was the work of people who believed their actions served a higher purpose."

The same construction appears in his tribute to his predecessor Bernhard Vogel:

"I am not merely stepping into a predecessor's office, but into the legacy of an aspiration."

The FAZ article raises a further suspicion: quotes from several experts may also have been invented by an AI tool. Language models are prone to hallucination, meaning they produce statements, sources or quotes that sound plausible but are made up. The report doesn't settle whether that actually happened here.

How the State Chancellery and the FAZ responded

The State Chancellery didn't reject the allegations. But it says AI systems are only used as an aid. There's an internal directive on using AI, it says, but it doesn't include a "general obligation to label texts that were created or supported using AI systems."

That wasn't enough for the FAZ. The paper only publishes AI-generated text in exceptional cases, and only with a label. Because the State Chancellery couldn't credibly show that the op-ed had been written without substantial AI help, the editors unpublished it. It's unusual for a major newspaper to pull a piece by a sitting head of government from its archive entirely for this reason.

Political context: Voigt was already under pressure

For about a year and a half, Voigt has led a so-called blackberry coalition of the CDU, BSW and SPD, which doesn't have its own majority in the Thuringian state parliament. He's also facing plagiarism allegations: after reviewing his dissertation, Chemnitz University of Technology revoked his doctorate. Voigt has filed an objection and says he'll go to court if necessary. So the new affair hits him at a time when his credibility as the author of his own texts was already in question.

The 49-year-old makes no secret of how he feels about AI. He told the Tagesspiegel:

"I encourage everyone in our administration to use every modern tool available for their work in 2026. AI has long been part of modern communication."

If individual passages were written with AI help, he added, he won't bite anyone's head off over it.

Our take

Politicians not writing their own speeches is nothing new. Every state chancellery has speechwriters. What's new is the question of whether it matters when a language model supplies the text, especially for occasions as sensitive as Holocaust remembrance. Many people will have a problem with words of remembrance coming from a machine that strings them together based on statistical patterns. The suspicion of invented expert quotes is even more serious. At that point it's no longer about style but about truth.

That said, caution is in order. AI detectors like Pangram give you probabilities, not proof. They can wrongly flag human-written text as AI-generated, especially formulaic language of the kind you often find in political speeches. The most solid fact here is that the State Chancellery admits to using AI in principle. How much AI was actually involved in each case remains unclear.

The case exposes a gap that probably affects many public authorities: governments are using AI more and more, but clear labeling rules are often missing. Outlets like the FAZ have set their own standards, and those don't necessarily match what authorities allow internally. Expect this case to fuel the debate over whether public statements by officials should be labeled when AI was involved.