Political Bias in AI Voting Tools
How LLMs exhibit concerning left-wing bias in election information
This study reveals that AI-based voting advice tools demonstrate significant political biases when providing supposedly neutral information to voters.
Key findings:
- LLMs show over 75% alignment with left-wing party positions in German political contexts
- AI systems exhibit vulnerability to prompt injections, leading to hallucinations
- The tools present concerning risks to electoral integrity when positioned as neutral information sources
Security implications: As AI tools become increasingly prevalent in election cycles, these biases could undermine democratic processes by presenting skewed political information as objective, potentially influencing voter decisions through subtle manipulation.