Geographic Bias in LLM Fact-Checking
Uncovering performance disparities across global regions
This research evaluates how LLMs perform at fact-checking across different geographic regions, revealing critical security implications for misinformation detection.
- Performance varies significantly across different global regions
- Testing of both open and closed-source models across 600 fact-checked statements
- Evaluation of three fact-checking approaches: basic LLM, agent-based, and retrieval-augmented
- Findings highlight security vulnerabilities in global information verification systems
Understanding these regional disparities is crucial for developing more equitable and reliable AI-based fact-checking tools to combat misinformation threats worldwide.