优先展示覆盖情况与实测结果；需要审计时再打开运行记录和技术细节。
0/4
条结论获得证据支持
0
获得支持
0
受到挑战或冲突
0
遭到反驳
1
无法判定
3
尚未评估
每一行是一条真实执行记录；命令、日志、哈希和签名都收纳在详情中。
逐个实验展示当前执行状态、阻塞原因、恢复动作和证据去向；技术执行与科研结论始终分开。 另有 1 条结论本次未安排独立复现。
执行成功不等于论文结论成立
目标状态回答平台有没有跑完；右侧科学结论只由不可变证据和 Assessment 决定。资源不足或平台故障不会被写成反驳论文的科研结论。
Prevention-based defenses that perform well on short-context benchmarks retain substantial attack success on the paper's long-context synthetic benchmark, although PromptLocate and MetaSecAlign 8B are lower than several simpler defenses on some tasks.
这条结论尚无可执行实验方案。
The paper reports that detection-based defenses exhibit an extreme false-positive/false-negative trade-off on long-context inputs, with some methods flagging benign inputs and others missing attacks.
这条结论尚无可执行实验方案。
The paper reports that GCG and its universal variant achieve high attack success across all four task suites and outperform heuristic attacks on several tasks.
这条结论尚无可执行实验方案。
On long-context document tasks, heuristic prompt-injection attacks substantially increase attack success rate over no-attack baselines, and Authority spoof is generally the strongest listed heuristic.
Independent approximate paper-review Authority spoof ASR on controlled long contexts
下一步
执行链路已完成；继续查看科学 Assessment。
结论与实验绑定已建立
证据已发布
证据发布 · 已完成
1 个不可变 Artifact
已有评估但无法判定
历史任务未记录目标级资源要求
