AI arrives in Parliament: What Sweden’s chatbot lawmakers could mean for Canada

Artificial intelligence is finding its way into almost every profession, from software development and finance to healthcare and education. Now a new study suggests that politicians are increasingly joining the trend, raising important questions about transparency, public trust and the future of democratic decision-making.

Researchers from Sweden’s Chalmers University of Technology have discovered that a significant proportion of parliamentary motions submitted in Sweden and the United Kingdom appear to contain text generated or assisted by large language models (LLMs) such as ChatGPT. Their findings suggest that AI is quietly becoming part of the political drafting process, even though politicians are rarely, if ever, disclosing its use. The research, published as the preprint study “Detecting undisclosed LLM-generated content in parliamentary texts”, found that during Sweden’s most recent parliamentary session, almost one in ten political motions contained AI-assisted content. The researchers report that none of the politicians involved disclosed that AI had been used.

While the study focused on Sweden and the UK, its implications extend far beyond Europe. Canada, which is positioning itself as a global leader in artificial intelligence, may soon face similar debates about transparency, accountability and the appropriate use of AI in public office.

Politicians are already using AI

The researchers developed bespoke AI-detection systems trained specifically on parliamentary language. Rather than relying on generic AI detectors, which can be unreliable, the team built models using thousands of parliamentary motions written before modern AI chatbots existed. To create a comparison set, they asked AI systems to rewrite parliamentary motions based on summaries of original documents. By comparing genuine human-authored texts with AI-generated versions, they created detectors capable of identifying AI-related linguistic patterns with a high level of accuracy.

The results showed a rapid increase in the apparent use of AI assistance. In Sweden, the proportion of motions containing AI-assisted content rose from almost zero in 2021–2022 to 9.4 per cent of motions during the 2025–2026 parliamentary year. Perhaps more striking than the use of AI itself was the complete absence of disclosure.

“Regardless of the extent to which AI has been used, there may be risks associated with a lack of transparency,” said lead author Minerva Suvanto of Chalmers University. “It is also a matter of trust from the recipient’s perspective.”

Why democratic transparency matters

The issue is not necessarily whether politicians should use AI. Drafting policy documents frequently involves analysing legislation, summarising evidence, reviewing consultation responses and producing complex reports. AI tools can help organise information and improve efficiency. The concern is whether readers deserve to know when AI has contributed to the final text.

Across journalism, academia and business, it is increasingly common to acknowledge the use of generative AI. Academic publishers have issued guidance on AI-assisted writing, while many media organisations now identify when AI-generated material contributes to published content. The researchers argue that similar considerations may apply to political communication. Their concerns reflect broader public anxieties surrounding artificial intelligence. Sweden’s SOM Institute found that 88 per cent of Swedes expressed concern that false information generated by AI could influence democratic elections.

Canada occupies a distinctive position in the AI landscape. The country is home to world-leading research centres including Mila – Quebec Artificial Intelligence Institute and the Vector Institute in Toronto. Canadian researchers have played a major role in the development of modern machine-learning techniques. At the same time, the federal government has actively promoted AI innovation while also exploring regulatory frameworks designed to encourage trustworthy adoption.

This creates an interesting paradox. Canada is investing heavily in AI-driven productivity and innovation, yet politicians themselves may eventually become significant users of the technology. The Swedish findings therefore raise a question that Canadian lawmakers may soon have to confront: should elected representatives be required to disclose when AI contributes to speeches, policy proposals, consultation responses or parliamentary motions?

One of the reasons transparency matters is that modern AI systems produce remarkably convincing prose. The researchers note that AI-assisted text is often highly polished, coherent and persuasive. While these characteristics may improve readability, they can also make errors harder to detect.

Anyone who has used generative AI extensively will recognise this challenge. AI systems frequently generate fluent and authoritative-sounding responses, even when some factual statements are incorrect or unsupported by evidence. If politicians rely heavily on AI-generated drafting without independently verifying content, there is a risk that inaccuracies could become embedded within policy discussions. The danger is not necessarily malicious intent. It may simply reflect overconfidence in an apparently well-written document.

The timing of the study is particularly relevant because AI transparency regulations are evolving rapidly. On 2 August 2026, new provisions under the European Union’s AI Act came into force. The legislation introduces transparency requirements relating to AI-generated content and establishes obligations for certain users and providers of artificial intelligence systems.

The researchers emphasise that their work was not intended as a legal analysis of the new regulations. However, they suggest that politicians may need to reconsider how they report their use of AI if public trust is to be maintained. Canada is also developing its own approach to AI governance. Although the regulatory landscape differs from the European Union, the broader issues of transparency, accountability and trust remain highly relevant.