Safeguarding Large Language Models in Real-time with Tunable Safety-Performance Trade-offs

Large Language Models (LLMs) have been shown to be susceptible to jailbreak attacks, or adversarial attacks used to illicit high risk behavior from a model. Jailbreaks have been exploited by cybercriminals and blackhat actors to cause significant harm, highlighting the critical need to safeguard widely-deployed models. Safeguarding approaches, which include fine-tuning models or having LLMs "self-reflect", may lengthen the inference time of a model, incur a computational penalty, reduce the semantic fluency of an output, and restrict ``normal'' model behavior. Importantly, these Safety-Performance Trade-offs (SPTs) remain an understudied area. In this work, we introduce a novel safeguard, called SafeNudge, that combines Controlled Text Generation with "nudging", or using text interventions to change the behavior of a model. SafeNudge triggers during text-generation while a jailbreak attack is being executed, and can reduce successful jailbreak attempts by 30% by guiding the LLM towards a safe responses. It adds minimal latency to inference and has a negligible impact on the semantic fluency of outputs. Further, we allow for tunable SPTs. SafeNudge is open-source and available through https://pypi.org/, and is compatible with models loaded with the Hugging Face "transformers" library.

Key Contributions

SafeNudge: a real-time safeguard combining Controlled Text Generation with 'nudging' (text interventions) that activates mid-generation when a jailbreak attempt is detected
Reduces successful jailbreak attempts by 30% with minimal inference latency overhead and negligible impact on semantic fluency
Tunable Safety-Performance Trade-offs (SPTs) allowing practitioners to explicitly balance safety strictness against model utility