Electoral Hallucinations: Safeguarding UK Elections in the World of LLMs and AI Chatbots

Hancock and Moores's May 2026 report investigates a category of AI-generated election disinformation that receives less attention than deepfakes but potentially reaches more people: LLM hallucinations about electoral processes — polling station locations, registration deadlines, voter eligibility requirements, candidate positions — produced by AI chatbots responding to genuine voter queries. The report tests major AI chatbots against a battery of electoral questions and documents both the frequency and the plausibility of hallucinated responses.

The policy implication is important: unlike deliberately produced disinformation, AI hallucinations are a systemic failure of the information infrastructure itself — produced not by adversarial intent but by the inherent limitations of current language models. The counter-measures required are correspondingly different: model audit requirements to assess electoral accuracy, platform labelling obligations distinguishing AI-generated from authoritative electoral information, and proactive provision of verified electoral information through channels that AI systems draw on for training. For election officials, regulators, and AI governance practitioners, this report identifies a genuinely underappreciated risk.