This study evaluates how eight open-source large language models respond to prompts containing deliberate misinformation presented with varying levels of confidence.

• Models struggle most with strongly confident misinformation, highlighting vulnerability to authoritative-sounding falsehoods
• Performance varies significantly across models, with newer architectures showing greater resilience
• When misinformation is presented with limited confidence, models demonstrate improved ability to identify and reject false assertions
• Findings reveal critical security implications for LLM deployment in high-stakes domains

This research provides essential insights for organizations implementing LLMs in environments where misinformation poses security threats, offering a benchmark for model selection and security protocols.

Battling Misinformation: An Empirical Study on Adversarial Factuality in Open-Source Large Language Models