← all datasets

BELLS

Emerging
1papers using it
2025first seen

BELLS is a Benchmark for the Evaluation of LLM Supervision Systems that contains a dataset covering 3 jailbreak families and 11 harm categories, used to evaluate the effectiveness of supervision systems against diverse attacks with varying harm severity and adversarial sophistication.

Papers using BELLS (1)

BELLS β€” datasets β€” cybersecurity