An investigation by Dutch outlets Investico and De Groene Amsterdammer found that AI detection tool Pangram classified 297 parliamentary contributions as mostly AI-written, bringing scrutiny to lawmakers’ use of AI in public debate.
The outlets published the investigation Oct. 21. Their review covered prepared opening remarks at parliamentary and committee debates held Jan. 1 through Sept. 30, 2026. Speeches by 68 people who served as Dutch lawmakers during the period included at least one contribution that met the test. The group includes former House member Esmah Lahlah.
Researchers counted a contribution only when Pangram assessed it as more than 50% AI-written with high confidence. They reviewed 2,715 contributions, and found at least one qualifying item in 187 of the 303 debates. Interruptions and motions were excluded.
The findings describe the detector’s assessment of submitted text. They do not establish how much a lawmaker wrote before using AI or prove who composed a particular passage. Because the study covered prepared debate openings, its count does not represent every spoken or written contribution by lawmakers during the period.
None of the lawmakers confronted with the findings denied using AI, the investigation said. Several said they had used it to shorten or refine text they drafted themselves.
BBB parliamentary leader Henk Vermeer told the investigators that AI tools help a small party manage a heavy workload. Volt leader Laurens Dassen told them that deciding what to cut from a speech remains part of a lawmaker’s judgment, even when AI is used to shorten it.
Five of the House’s 17 party groups had firm agreements on AI use, the investigation found. The CDA was developing guidelines. The outlets also cited an earlier NOS study of AI-assisted parliamentary questions, a separate area of legislative work.
