超越指令覆盖：用任务篡改攻击对间接提示注入进行基准测试
Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacks AI Safety HOT (aisafetyhot.com), CC BY-NC 4.0 · 本地公开归档导入
推荐理由：用户指定的本地展示归档；未运行评分模型，未生成分数。
Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacks AI Safety HOT (aisafetyhot.com), CC BY-NC 4.0 · 本地公开归档导入
推荐理由：用户指定的本地展示归档；未运行评分模型，未生成分数。