With the task-only system message and objective A, defenses had little effect on naive and lookalike injections; nameless tokenization reduced forged-turn success from 99.9% to 92.8% and forged-tool success from 33.1% to 7.4%, but increased forged-system success from 46.9% to 68.0%.

With the task-only system message and objective A, defenses had little effect on naive and lookalike injections; nameless tokenization reduced forged-turn success from 99.9% to 92.8% and forged-tool success from 33.1% to 7.4%, but increased forged-system success from 46.9% to 68.0%. · CiteArk

测量结果

已评估 0/29 项

获得支持 0有争议或冲突 0尚未评估 29

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
attack: Naive · defense: None	62.5%	—	尚未评估
attack: Naive · defense: Strip	62.2%	—	尚未评估
attack: Naive · defense: Mask	62.2%	—	尚未评估
attack: Naive · defense: Escape	62.3%	—	尚未评估
attack: Naive · defense: Nameless	62.4%	—	尚未评估
attack: Lookalike · defense: None	93%	—	尚未评估
attack: Lookalike · defense: Strip	93%	—	尚未评估
attack: Lookalike · defense: Mask	93%	—	尚未评估
attack: Lookalike · defense: Escape	93%	—	尚未评估
attack: Lookalike · defense: Nameless	93%	—	尚未评估
attack: Forged turn · defense: None	99.9%	—	尚未评估
attack: Forged turn · defense: None	99.8%	—	尚未评估

With the task-only system message and objective A, defenses had little effect on naive and lookalike injections; nameless tokenization reduced forged-turn success from 99.9% to 92.8% and forged-tool success from 33.1% to 7.4%, but increased forged-system success from 46.9% to 68.0%.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果