arXiv cs.CL· Wenhui Chu·· 12 小时前AI 评分42
研究：LLM 智能体的交接摘要何时变成指令？自写摘要中出现无依据的策略限制
When Do Summaries Become Instructions? Unsupported Policy in LLM Agents' Self-Written Handoff Summaries
AI 导读
研究考察长时运行 LLM 智能体的交接摘要是否会写入记录不支持的限制或授权。两个开放权重模型的可读摘要中，无依据限制分别占 1.2% 和 2.4%，无依据授权占 0.1% 和 1.3%；将虚假授权中和后，120 次留出运行中的未授权操作从 24 次降至 7 次。
正文 · 原文
Abstract:Long-running LLM agents compact their context, replacing most history with a summary. We ask whether handoff summaries state permissions or requirements the record does not support, whether such sentences change what the next instance (the successor) does, and where they come from. Synthetic tool-use tasks carry a ground-truth permission ledger; same-snapshot replay reruns the successor with those whole sentences replaced by neutral filler. Two AI readers, reading only rule-flagged held-out summaries, confirmed unsupported restrictions in 1.2% and 2.4% of all readable summaries from two open-weight models, and unsupported authorizations in 0.1% and 1.3%. For restrictions, a pre-specified replay test was not significant (p = 0.58); a follow-up, planned after it, found a fragile effect (p = 0.027). Neutralizing false approvals cut unauthorized actions from 24 to 7 of 120 held-out runs (post hoc; later set: 17 to 11 of 48). Most development restrictions followed the agent's earlier reasoning that an action needs approval (post hoc). In pre-specified tests, replacing the system prompt's "Some actions need approval." with one saying where requirements come from cut unsupported restrictions from 5.0% to 1.8% of development summaries and, on held-out templates, from 1.8% to 0.3% (both p < 10^-9), while requests for approvals the ledger requires barely changed. Code and data: this https URL
| Comments: | 22 pages, 3 figures, 15 tables. Code and data: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.06545 [cs.CL] |
| (or arXiv:2611.06545v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.06545 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wenhui Chu [view email]
[v1]
Wed, 4 Nov 2026 14:12:09 UTC (127 KB)
来源：arXiv cs.CL · arxiv.org