Five Official Posts Found: Dutch PM Told Parliament No AI Was Used
October 9, 2026 · Frisian News
Photo: Wikimedia Commons
Prime Minister Rob Jetten told parliament on September 29 that no AI was used on his official accounts. De Volkskrant then identified five AI-generated posts on those same accounts, directly contradicting his written answer.
Five. That is the number of official posts from the Dutch Prime Minister's MinPres accounts that De Volkskrant identified as machine-generated, days after Prime Minister Rob Jetten told parliament in a written answer on September 29 that "no use was made of AI" on those accounts.
The paper ran the MinPres feed through Pangram, the AI detection model it had used in its September investigation, and found five posts averaging around 200 words each flagged as generated. As a check, De Volkskrant also ran 700 posts Jetten and his party published before ChatGPT's release in 2022 through the same model. None was flagged.
The margin for error is slim. Researchers at the universities of Chicago and Maryland found Pangram wrongly identifies a genuine human text as machine-written in about one case in 10,000. Two language experts asked by De Volkskrant to assess the posts said five false positives at that length was not a serious possibility. AI researcher Marijke Van Vlasselaer told the paper the posts shared "the same cadence and structure every time," opening with phrases such as "The Netherlands and Germany are neighbours," "The world is restless," and "Everyone knows it."
Jetten had drawn a careful line in his September 29 parliamentary reply, separating his private accounts from his official MinPres handle and telling parliament that AI use on official channels was zero. The Government Information Service (RVD) backed that position, telling De Volkskrant that AI had been used exclusively on private accounts managed by Jetten and D66, meaning the government's 2025 guidelines would not apply. Those guidelines prohibit verbatim copying of AI text and require disclosed labelling.
The RVD has since shifted ground. After De Volkskrant presented the five flagged posts, the service said it does not deny AI was used but "does not share the conclusion that the cited examples were generated entirely by AI." It pointed to policy dossiers, meeting notes, and previously used text as possible explanations. It also directed the paper to a newer, looser rule that does not require disclosure of AI use provided a human has reviewed the content. According to De Volkskrant, website metadata shows that guideline was created on September 17, one week after the newspaper's first story on September 10.
Wim Voermans, professor of constitutional law at Leiden University, said Jetten's written answer left no political room. "If you state flatly that AI was not used, and it can be shown that it was, that cannot be explained away politically." Jetten told RTL Nieuws: "You put something in, you take something out, and you always check it," a description that does not match a categorical denial to parliament.
The five official posts sit inside a larger pattern. De Volkskrant's September 10 report found at least 232 posts across Jetten's personal accounts and D66's party feeds were machine-written: 68 in Jetten's own name and 164 from the party. From June 4, 2025, the day after the Schoof cabinet collapsed, roughly 30 per cent of all posts from Jetten and D66 were AI-generated, covering subjects including Ukraine, the MH17 commemoration, and the government's apology to the Moluccan people. The leaders of the other two coalition parties, Dilan Yesilgöz of VVD and Henri Bontenbal of CDA, showed virtually no AI-generated posts in the same analysis.
When the evidence mounted, the government changed the definition of disclosure, not the policy.
Published October 9, 2026 · Frisian News