2 matches found
StealthRL
StealthRL:用于多检测器规避 AI 文本检测器的强化学习改述攻击 论文(arXiv) 演示 模型(Hugging Face) 基准数据集(Hugging Face) 摘要 AI 文本检测器越来越多地用于高风险场景,但它们对保持语义的对抗性改写的鲁棒性仍不确定。我们提出了 StealthRL,一个通过自适应改述攻击对 AI 文本检测器进行压力测试的强化学习框架。StealthRL 针对检测器集成训练改述策略,同时保留语义内容,然后评估对留出检测器家族的迁移性。在完整过滤后的 MAGE 测试池(15,310 条人类样本 / 14,656 条 AI 样本)上,StealthRL 将平均...
6AI score
SaveExploits0
Fooling the Watchers: Breaking AIGC Detectors Via Semantic Prompt Attacks
The rise of text-to-image T2I models has enabled the synthesis of photorealistic human portraits, raising serious concerns about identity misuse and the robustness of AIGC detectors. In this work, we propose an automated adversarial prompt generation framework that leverages a grammar tree...
6.9AI score
SaveExploits0
20