FrankensteinBench

Emerging

1papers using it

2026first seen

'FrankensteinBench' is a safety benchmark containing 11,279 malicious queries, categorized by complexity, used to evaluate the effectiveness of jailbreak strategies for large language models.

🔎 Find this dataset

Papers using FrankensteinBench (1)

Jailbreaking for the Average Jane: Choosing Optimal Jailbreaks via Bandit Algorithms for Automatically Enhanced Queries2026