The UK Electoral Commission has called for new legal controls over misinformation from AI chatbots after a thinktank discovered they made serious factual errors during the recent Scottish elections.
Thinktank Demos revealed that AI services provided voters with misinformation in 34% of the responses to questions posed, raising urgent questions about the lack of regulation for AI platforms in the UK. Ahead of May's Holyrood election, Demos simulated voter queries by submitting 75 questions to five free AI tools—including ChatGPT, Google Gemini, and Replika—focusing on three real-life constituencies to evaluate their accuracy and evidence-based reliability.
In its report, titled "Electoral Hallucinations," Demos reported that the tested AI tools invented fictitious financial scandals, provided wrong election dates, falsely claimed Scottish voters needed ID at polling stations, and incorrectly assigned candidates to wrong contests.
An accompanying poll of 2,005 British adults commissioned by Demos indicated that 20% of voters (equivalent to roughly 10 million people UK-wide) used AI chatbots or search tools to gather information on parliamentary elections in Scotland, Wales, and English local councils.
Vijay Rangarajan, Chief Executive of the Electoral Commission, has been pressing government ministers to introduce legislation holding AI corporations accountable, following findings that half of the voters in the 2024 general election encountered misleading information. "Voters want accurate information to help them engage with democracy, and it is concerning that AI tools have made the spread of false or misleading information dramatically faster and more accessible than ever," he stated. "The current legal framework should go further."
ChatGPT delivered wrong information in 46% of its answers, which included fabricating a parliamentary expenses scandal. Rangarajan argued that ministers should introduce clearer statutory duties on AI platforms to protect voters from misinformation, giving media regulator Ofcom stronger enforcement powers.
Azzurra Moores, Associate Director at Demos, emphasized the global nature of this threat: "The accessibility of these AI tools—all developed and run by US corporations—is widespread in the UK, but we don't yet have the legislative framework to protect our democracy from the knock-on impact of its circulation." She urged ministers to make AI companies liable under UK defamation and electoral laws, establish mandatory accuracy safeguards, and force AI firms to allow independent researchers to audit their internal data and training methodologies.
According to Demos, the companion chatbot Replika performed the worst, failing in 56% of its answers by inventing dates for fictitious expenses scandals, fabricating candidates, and making up groundless accusations.
[AgentUpdate Depth Analysis] As AI Agents transition into critical daily workflows, factual grounding and real-time alignment are becoming non-negotiable baselines for trust. The widespread hallucinations during the Scottish election highlight a fundamental flaw of raw LLMs in high-stakes, time-sensitive domains. Because users interact with Agents as authoritative, conversational guides, the societal impact of their misinformation is significantly magnified compared to traditional search engines. For the Agent ecosystem to mature, developers must move beyond single-prompt generation toward multi-agent verification frameworks and advanced Retrieval-Augmented Generation (RAG). Building robust semantic 'guardrails' and implementing cross-source consensus verification within agentic workflows is essential. To unlock enterprise and civic adoption, the industry must treat deterministic fact-checking not as an afterthought, but as a core layer of the Agent architecture.
