PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flips
This reveals a critical security flaw in commercial LLMs, enabling harmful outputs without input modifications, which is a significant threat for AI safety and deployment.
The authors tackled the vulnerability of safety-aligned large language models (LLMs) to jailbreaking by flipping only 5 to 25 bits in model parameters, achieving attack success rates of 80-98% on 10 open-source LLMs with minimal utility impact.
We study a new vulnerability in commercial-scale safety-aligned large language models (LLMs): their refusal to generate harmful responses can be broken by flipping only a few bits in model parameters. Our attack jailbreaks billion-parameter language models with just 5 to 25 bit-flips, requiring up to 40$\times$ fewer bit flips than prior attacks on much smaller computer vision models. Unlike prompt-based jailbreaks, our method directly uncensors models in memory at runtime, enabling harmful outputs without requiring input-level modifications. Our key innovation is an efficient bit-selection algorithm that identifies critical bits for language model jailbreaks up to 20$\times$ faster than prior methods. We evaluate our attack on 10 open-source LLMs, achieving high attack success rates (ASRs) of 80-98% with minimal impact on model utility. We further demonstrate an end-to-end exploit via Rowhammer-based fault injection, reliably jailbreaking 5 models (69-91% ASR) on a GDDR6 GPU. Our analyses reveal that: (1) models with weaker post-training alignment require fewer bit-flips to jailbreak; (2) certain model components, e.g., value projection layers, are substantially more vulnerable; and (3) the attack is mechanistically different from existing jailbreak methods. We evaluate potential countermeasures and find that our attack remains effective against defenses at various stages of the LLM pipeline.