Artificial intelligence is increasingly finding its way into parliamentary speeches and legislative texts without voters necessarily being told, according to new research examining political writing in Sweden and the United Kingdom.
Euronews reported that researchers from Chalmers University of Technology and Edinburgh Napier University analyzed 13,565 Swedish parliamentary motions and 4,209 British parliamentary statements published between 2021 and April 2026. They used AI detectors trained to distinguish human political writing from text produced by large language models.
The results suggest adoption has accelerated quickly. Around 15% of the UK parliamentary statements examined in 2026 contained at least one paragraph classified as AI-generated. In Sweden, 9.4% of motions submitted during the 2025–2026 parliamentary year contained at least one paragraph classified as AI-assisted—roughly 300 motions.
AI use rose sharply, but disclosure did not
The researchers found no indication that politicians had disclosed AI use in the texts identified by their system.
Euronews said the proportion had risen considerably from earlier years, when approximately 3.5% of British texts in 2022 and 1.7% of Swedish texts in 2022–2023 showed signs of AI assistance. When researchers assessed entire documents rather than individual passages, 6.5% of Swedish motions in the latest parliamentary year and 2.1% of British texts in 2026 were classified as AI-generated.
The researchers also tested their detection system against parliamentary material from 2021, before ChatGPT became publicly available, and falsely classified only one complete text in each country as AI-generated. Still, the study could not determine which AI tools politicians used or precisely how those systems contributed to individual texts.
Mattias Wahde, professor of applied artificial intelligence at Chalmers, said the researchers used explainable AI methods that allowed them to identify where AI-like patterns appeared within a document rather than simply assigning a single classification to the entire text.
Swedish Green Party parliamentarian Jan Riise, the party’s spokesperson on AI and digitalization, shared that his party uses AI for research purposes and has an internal policy encouraging caution. He also supported clearer labeling when politicians use the technology.
AI’s political problem extends beyond writing
The disclosure issue arrives alongside another problem: AI systems themselves can provide unreliable information about democratic processes.
Election Law Blog highlighted research finding that leading chatbots produced inaccurate, unclear or outdated answers to 29% of basic election and voting questions in testing ahead of the 2026 U.S. midterms. Performance was particularly weak when questions were asked in Spanish.
Together, the findings reveal two different transparency problems. Politicians may increasingly use generative systems to produce public-facing political material without disclosing that assistance, while citizens may turn to the same class of technology for election information that is not consistently reliable.
Political AI is becoming an infrastructure problem
For AI builders, this is more than a debate over whether politicians should use ChatGPT to polish speeches.
Political applications create unusually high requirements for provenance, disclosure, accuracy and auditability. A system used to brainstorm privately poses different risks from one contributing language to legislation or giving someone instructions on how to vote.
The studies therefore point toward a product challenge the AI industry has not fully solved: determining when generated content should carry provenance signals and when high-stakes answers should be anchored to authoritative sources.
As generative AI moves deeper into government workflows, transparency may need to become part of the technical architecture rather than a voluntary note added after the text has already been written.
