跳到正文
arXiv cs.AI· Jiankai Jin, Xiangzheng Zhang, Zhao Liu, Wenzhuo Xu, Dongdong Yang, Deyue Zhang, Quanchen Zou·· 10 小时前AI 评分48

ElephantAgent:智能体系统中的上下文状态连续性

ElephantAgent: Contextual State Continuity in Agentic Systems

AI 导读

研究者提出 ElephantAgent 协议,通过对智能体上下文中安全关键子集强制"上下文状态连续性"检查,防御工具与记忆投毒攻击。集成该协议的 ElephantDSH 抵御了全部带外攻击(含 rug pull、回滚及针对自身机制的攻击),并在所有试验中从带内提示词注入中恢复;而评估的 5 种现有防御在回滚攻击下全部泄露目标内容,其中 3 种在 rug pull 下也泄露。

正文 · 原文

View PDF HTML (experimental)

Abstract:Agentic systems enhance their capabilities by invoking external tools and maintaining persistent memory. However, recent tool and memory poisoning attacks can covertly bias agent behavior. We observe that such attacks mounted out of band share a single enabler: each succeeds only by presenting the agent with a contextual state (e.g., tool state and memory) that is not the latest faithfully evolved one. This observation suggests that a continuity check over the contextual state covers this family of attacks. We present ElephantAgent, a protocol that enforces Contextual State Continuity over the bounded, security-critical subset of the agent's context. Before each planning-and-execution cycle, ElephantAgent recomputes the digest of the local contextual state and verifies it against the latest authorized digest in a linearizable ledger of state transitions, maintained over replicated trusted hardware. The protocol also provides Historical Traceability, a secondary guarantee that keeps state transitions attributable and recoverable. We integrate ElephantAgent into DeepSeek Harness (DSH) and implement ElephantDSH, an agentic system protected by contextual state verification. ElephantDSH withstood all out-of-band attack attempts, from rug pulls and rollbacks to attacks on its own mechanism, and recovered from an in-band prompt injection in all trials. By comparison, existing defenses typically check only what enters the context: of the five we evaluated, all leaked the targeted content under the rollback, and three also under the rug pull. ElephantDSH completed all normal coding tasks with a relative overhead of 1.02-1.07$\times$ over the original DSH with model latency variation removed.
Subjects: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
Cite as: arXiv:2607.01919 [cs.AI]
  (or arXiv:2607.01919v2 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2607.01919

arXiv-issued DOI via DataCite

Submission history

From: Jiankai Jin [view email]
[v1] Thu, 2 Jul 2026 09:18:09 UTC (1,005 KB)
[v2] Wed, 18 Nov 2026 08:55:08 UTC (1,197 KB)

来源:arXiv cs.AI · arxiv.org