# 自主智能体安全无法跨迭代组合

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-08-31 11:29
- AIHOT 分数：51
- AIHOT 链接：https://aihot.virxact.com/items/cmtgp7kyu05karoch7ej3tsqh
- 原文链接：https://x.com/rohanpaul_ai/status/2094266174548705644

## AI 摘要

新论文《Safety Does Not Compose》指出，自主智能体若跨迭代记忆，其安全系统就不能在迭代间遗忘，长期任务安全无法简化为重复运行安全的短期轨迹。攻击者可将恶意证据分散在多个看似良性的步骤中，仅基于轨迹的监控器无法积累足够上下文识别攻击。论文地址：arxiv.org/abs/2608.27141。

## 正文

If an autonomous agent remembers across iterations, its safety system cannot afford to forget between them..

long-horizon agent safety cannot be reduced to repeatedly running a safe short-horizon trajectory.

An attacker can split malicious evidence across several individually benign-looking steps, so a trajectory-only monitor never sees enough context to distinguish the attack from normal work.

– arxiv. org/abs/2608.27141

Title: "Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents"