用 ASCII 艺术规避毒性检测:针对审核系统的空间攻击基准
Evading Toxicity Detection with ASCII-art: A Benchmark of Spatial Attacks on Moderation Systems
AI总结:
本研究提出利用 ASCII 艺术的空间结构文本攻击毒性检测模型,构建 ToxASCII 基准,并在多种先进模型和审核工具上实现完美攻击成功率,揭示文本审核系统的重大漏洞。
摘要:
我们提出了一类新颖的对抗性攻击,针对毒性检测模型,利用语言模型在解释以 ASCII 艺术形式呈现的空间结构文本时的失败。为了评估这些攻击的有效性,我们提出了 ToxASCII,一个旨在评估毒性检测系统对视觉混淆输入鲁棒性的基准。我们的攻击在多种最先进的大型语言模型和专用审核工具上实现了完美的攻击成功率(ASR),揭示了当前仅基于文本的审核系统中的一个重大漏洞。
英文摘要:
We introduce a novel class of adversarial attacks on toxicity detection models that exploit language models' failure to interpret spatially structured text in the form of ASCII art. To evaluate the effectiveness of these attacks, we propose ToxASCII, a benchmark designed to assess the robustness of toxicity detection systems against visually obfuscated inputs. Our attacks achieve a perfect Attack Success Rate (ASR) across a diverse set of state-of-the-art large language models and dedicated moderation tools, revealing a significant vulnerability in current text-only moderation systems.
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。