Yongtaek Lim

Financial LLMs can now be rigorously evaluated against targeted risks, reducing critical false negatives in safety assessments from 28 to 12.

Culturally-adapted prompts can boost LLM safety evaluations by revealing a 9.3 percentage point increase in attack success rates compared to direct translations.

Mapping LLM attack strategies onto a multiplex network reveals interpretable vulnerability clusters and dramatically improves red teaming efficiency.