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

2025-01-02

AI Summary

Large language models (LLMs) remain vulnerable to jailbreak attacks, posing a fundamental trade-off between safety and performance. To address this, we propose SafeNudge—a real-time, lightweight inference-time defense mechanism. SafeNudge integrates controllable text generation with dynamic, context-aware textual “nudging” to detect and block adversarial inputs during inference, implemented seamlessly within the Hugging Face Transformers framework. Its core innovation is the first tunable Safety–Performance Trade-off (SPT) mechanism, enabling fine-grained configuration of security strength without incurring significant latency overhead (<1%) or degrading semantic fluency (BLEU score reduction <0.3). Extensive experiments demonstrate that SafeNudge reduces jailbreak success rates by 30% across diverse attack vectors. The method is open-sourced, compatible with mainstream open-weight LLMs, and deployable as a plug-and-play PyPI package—offering practical, low-overhead robustness for production LLM applications.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

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.

Problem

• Large Language Models
• Escape Attack Defense
• Performance-Security Tradeoff

Innovation

• SafeNudge
• LanguageModelSecurity
• PerformanceBalance