News
UK and Swedish Politicians Use AI to Write Speeches and Bills Without Telling Voters
A study from Chalmers University of Technology and Edinburgh Napier University found that up to 15 percent of UK parliamentary statements and 9.4 percent of Swedish parliamentary motions contain AI-written passages, none of them disclosed to voters.
Contents
Researchers from Chalmers University of Technology in Sweden and Edinburgh Napier University in the UK have found that politicians in both countries are increasingly using artificial intelligence to write speeches, motions and other parliamentary documents without telling voters. In none of the analyzed texts did the researchers find any disclosure that AI had been involved.
How Researchers Detected the AI
The team, led by Chalmers University of Technology PhD student Minerva Suvanto, built separate detectors for Swedish and English texts. The models were trained on two datasets: documents written by humans before the era of generative AI, and versions of the same topics generated by language models. This allowed the researchers to distinguish typical human style from the telltale patterns of machine-written text.
Co-author Mattias Wahde, a professor of applied artificial intelligence, stressed that the team used an approach known as interpretable AI, meaning AI whose decisions can be explained. This made it possible not only to determine whether a text was AI-assisted, but also to pinpoint the specific passages showing traces of machine writing.
Two Parliaments, Two Different Scales
In the UK, the increase is markedly faster than in Sweden. The share of statements with AI passages rose from 3.5 percent in 2022 to 15 percent in 2026, more than double the pace seen in the Riksdag, where the share of motions with AI passages rose from 1.7 to 9.4 percent over a similar period.
The gap between counting paragraphs and counting whole documents shows that in most cases politicians aren't handing the entire text over to a model, but weaving AI-generated passages into documents they otherwise wrote themselves. Even so, the researchers note that partial use still raises questions about transparency in the legislative process.
Regardless of the extent of AI use, there can be risks tied to a lack of transparency - Minerva Suvanto, PhD student, Chalmers University of Technology
The Risk of Unnoticed Errors
The study's authors warn that text produced by large language models is often very well written and persuasive, which paradoxically increases the risk that factual errors or inaccuracies go unnoticed, even by the politician who signs off on it. In legal documents such as parliamentary motions, such a mistake can translate into real legislative consequences.
Jan Riise, a member of parliament for Sweden's Green Party, weighed in on the study, saying the results didn't surprise him. His party uses AI mainly for research purposes and has internal guidelines urging caution. Riise came out in favor of mandatory labeling of AI-assisted texts, while cautioning that having a model write an entire motion is a very different situation from using it to support research.
If you have to write the whole motion, that's a different story - Jan Riise, MP, Swedish Green Party
Context From the Top
The study isn't an isolated signal. Swedish Prime Minister Ulf Kristersson revealed back in August 2025 that he regularly uses ChatGPT and the French LeChat to get a second opinion when making policy decisions. That shows AI tools have already become part of daily practice for top officials, not just rank-and-file parliamentarians.
For Polish readers, the study's relevance extends beyond the UK and Sweden. A debate is underway over rules for labeling AI-generated content in government and business, and Polish companies themselves admit they are struggling to navigate the regulations on labeling such content. The Swedish-British team's findings show that the mere absence of clear regulation doesn't stop the practice, it just keeps it invisible to the public.
The researchers recommend that parliaments and political parties introduce a requirement to openly label documents created with substantial AI involvement, similar to how many countries now require labeling of machine-generated video or graphics. For now, none of the parliaments studied has such a requirement, and the decision to disclose AI use remains entirely up to the individual politician.
