A research project I worked on Fall 2025. Accepted to ACL Findings 2026! About how current LLMs consistently stay faithful to provided context/scientific evidence, even when such evidence contains false and possibly harmful information (overriding safety), particularly in medicine (high stakes domain).