#MLSecOps
"Fine-Tuning Jailbreaks under Highly Constrained Black-Box Settings: A Three-Pronged Approach", 2025.
// Fine-tuning is a widely used method for adapting models to downstream tasks, yet it is vulnerable to jailbreak attacks. Our attack combines safety-styled prefix/suffix wrappers, benign lexical encodings (underscoring) of sensitive tokens, and a backdoor mechanism, enabling the model to learn harmful behaviors while individual datapoints appear innocuous
Post #1628
140