arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2604.13301cs.CRcs.AI
收藏

蜜罐协议

Honeypot Protocol

University of North Carolina at Pembroke(北卡罗来纳大学帕克维尔分校)

Najmul Hasan

AI总结:

针对AI控制中被动监控易受自适应攻击的问题,提出蜜罐协议,通过仅改变系统提示词测试模型上下文依赖行为,实验表明模型在各条件下均无异常行为。

摘要:

可信监控是AI控制中的标准防御手段,但容易受到自适应攻击、合谋和策略性攻击选择的影响。所有这些都利用了监控是被动的这一事实:它观察模型行为,但从不探测模型在不同感知条件下是否会表现出不同的行为。我们引入了蜜罐协议,该协议通过仅改变三种条件(评估、合成部署、明确无监控)下的系统提示词来测试上下文依赖行为,同时保持任务、环境和评分相同。我们在BashArena中评估了Claude Opus 4.6在诚实和攻击模式下所有三种条件下的表现。该模型在所有条件下均实现了100%的主任务成功率,并触发了零个侧任务,为未来与更强攻击策略和更多模型的比较提供了基线。

英文摘要:

Trusted monitoring, the standard defense in AI control, is vulnerable to adaptive attacks, collusion, and strategic attack selection. All of these exploit the fact that monitoring is passive: it observes model behavior but never probes whether the model would behave differently under different perceived conditions. We introduce the honeypot protocol, which tests for context-dependent behavior by varying only the system prompt across three conditions (evaluation, synthetic deployment, explicit no-monitoring) while holding the task, environment, and scoring identical. We evaluate Claude Opus 4.6 in BashArena across all three conditions in both honest and attack modes. The model achieved 100% main task success and triggered zero side tasks uniformly across conditions, providing a baseline for future comparisons with stronger attack policies and additional models.

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。

↑