Executive Summary
As Large Language Models (LLMs) become more integrated into everyday digital tools, countries around the world are grappling with both their potential and their risks. With the return of local elections in Sri Lanka after nearly seven years, and a new generation of voters increasingly turning to AI-powered chatbots and search engines for electoral information, the risk of automated misinformation takes on critical urgency. The integration of these models into mainstream platforms, combined with limited oversight and evolving regulatory frameworks, poses real threats to democratic participation and electoral integrity. This study examines these risks within the context of the 2025 Sri Lanka local elections, examining if and how LLM errors might misinform, exclude, or confuse the electorate. In this study, we tested four chatbots in English, Tamil, and Sinhala using 18 questions related to the electoral process and key political issues dominating the campaign in Sri Lanka.
