arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.01946cs.CRcs.CL
收藏

Bileve:利用双层签名保障大语言模型文本来源安全以抵御伪造攻击

Bileve: Securing Text Provenance in Large Language Models Against Spoofing with Bi-level Signature

Northeastern University(东北大学);UC Berkeley(加利福尼亚大学伯克利分校);UC Riverside(加利福尼亚大学河滨分校)

Tong Zhou, Xuandong Zhao, Xiaolin Xu, Shaolei Ren

AI总结:

针对现有大语言模型水印易受伪造攻击的问题,提出双层签名方案Bileve,通过基于排序的采样策略嵌入粗细两类签名,可区分5种检测场景,在OPT-1.3B和LLaMA-7B上验证了其抗伪造攻击的有效性。

摘要:

大语言模型(LLM)文本水印已被广泛用于识别机器生成内容的来源,在打击深度伪造或有害内容时的责任认定方面颇具前景。现有水印技术通常优先考虑抵御移除攻击的鲁棒性,但遗憾的是,它们易受伪造攻击:恶意行为者可以微妙地改变LLM生成回复的含义,甚至伪造有害内容,有可能将责任错误归咎于LLM开发者。为解决这一问题,我们提出了一种双层签名方案Bileve,该方案通过一种新颖的基于排序的采样策略,嵌入用于完整性检查的细粒度签名位(缓解伪造攻击),以及在签名无效时用于追踪文本来源的粗粒度信号(增强可检测性)。与仅输出二元结果的传统水印检测器相比,Bileve在检测过程中可区分5种场景,能够可靠地追踪文本来源并对LLM进行监管。在OPT-1.3B和LLaMA-7B上开展的实验证明,Bileve能够有效抵御伪造攻击,且具备更强的可检测性。代码可在https://github.com/Tongzhou0101/Bileve-official获取。

英文摘要:

Text watermarks for large language models (LLMs) have been commonly used to identify the origins of machine-generated content, which is promising for assessing liability when combating deepfake or harmful content. While existing watermarking techniques typically prioritize robustness against removal attacks, unfortunately, they are vulnerable to spoofing attacks: malicious actors can subtly alter the meanings of LLM-generated responses or even forge harmful content, potentially misattributing blame to the LLM developer. To overcome this, we introduce a bi-level signature scheme, Bileve, which embeds fine-grained signature bits for integrity checks (mitigating spoofing attacks) as well as a coarse-grained signal to trace text sources when the signature is invalid (enhancing detectability) via a novel rank-based sampling strategy. Compared to conventional watermark detectors that only output binary results, Bileve can differentiate 5 scenarios during detection, reliably tracing text provenance and regulating LLMs. The experiments conducted on OPT-1.3B and LLaMA-7B demonstrate the effectiveness of Bileve in defeating spoofing attacks with enhanced detectability. Code is available at https://github.com/Tongzhou0101/Bileve-official.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑