arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2410.03489cs.CRcs.AI
收藏

面向多模态融合模型的基于梯度的越狱图像

Gradient-based Jailbreak Images for Multimodal Fusion Models

Meta;ETH Zurich(苏黎世联邦理工学院)

Javier Rando, Hannah Korevaar, Erik Brinkman, Ivan Evtimov, Florian Tramèr

AI总结:

针对多模态融合模型分词不可微的问题,提出分词器捷径实现端到端梯度图像越狱攻击,并在 Chameleon 上验证其高效性及文本防御的可迁移性。

摘要:

为语言模型增加图像输入,可能通过连续优化实现更有效的越狱攻击,这不同于需要离散优化的文本输入。然而,新型多模态融合模型使用不可微函数对所有输入模态进行分词,这阻碍了直接攻击。在本工作中,我们引入了分词器捷径(tokenizer shortcut)的概念,用连续函数近似分词过程,从而实现连续优化。我们利用分词器捷径创建了首批针对多模态融合模型的端到端梯度图像攻击。我们在 Chameleon 模型上评估了这些攻击,得到的越狱图像能够在 72.5% 的提示词上诱导出有害信息。越狱图像优于使用相同目标优化的文本越狱,并且在优化多达 50 倍输入 token 时,所需计算预算降低至三分之一。最后,我们发现,仅在文本攻击上训练的表示工程防御(如 Circuit Breakers)能够有效迁移到对抗性图像输入。

英文摘要:

Augmenting language models with image inputs may enable more effective jailbreak attacks through continuous optimization, unlike text inputs that require discrete optimization. However, new multimodal fusion models tokenize all input modalities using non-differentiable functions, which hinders straightforward attacks. In this work, we introduce the notion of a tokenizer shortcut that approximates tokenization with a continuous function and enables continuous optimization. We use tokenizer shortcuts to create the first end-to-end gradient image attacks against multimodal fusion models. We evaluate our attacks on Chameleon models and obtain jailbreak images that elicit harmful information for 72.5% of prompts. Jailbreak images outperform text jailbreaks optimized with the same objective and require 3x lower compute budget to optimize 50x more input tokens. Finally, we find that representation engineering defenses, like Circuit Breakers, trained only on text attacks can effectively transfer to adversarial image inputs.

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。

↑