Overview
- On Wednesday the investigative portal Frag den Staat published an analysis that prompted the Frankfurter Allgemeine Zeitung to depublish a Mario Voigt guest article after the Staatskanzlei’s general reply failed to verify whether the text used generative AI.
- Frag den Staat used detection tools such as Pangram and GPTZero and reported high AI-probability signals across many texts, saying multiple speeches had over 50 percent AI share and three guest pieces showed 100 percent AI in the analysis.
- The Thüringer Staatskanzlei acknowledged that AI is used “supportively” to prepare speeches and texts but did not provide specific, verifiable answers about the passages and quotes questioned by the investigation.
- Investigators flagged passages that attribute direct quotes to named scientists that could not be verified and one cited researcher denied writing the attributed sentence, illustrating risks that generative models can produce fabricated attributions.
- The case sharpens political pressure on Voigt, who lost his doctorate earlier this year, spotlights the Muster‑Dienstanweisung’s disclosure rules and highlights uncertainty about detector accuracy and how upcoming EU AI Act transparency rules will apply.