跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.11506 · v1

On the Relative Security of Prompt Injection: Mechanised Definitions, Defences and Verdicts for LLM Agents

首次提交 2026-11-09 01:20
版本更新 2026-11-09 01:20
雷达观测 2026-11-10 13:19
arXiv ↗PDF ↗

摘要

An LLM agent that reads untrusted content, such as an email or a web page, may follow instructions hidden in it. Noninterference is too strong for such agents, since a useful agent lets untrusted content influence what it does and may disclose the user's data by design, and declassification needs a release policy written for each task. We instead state security relative to a reference agent, which runs the same plan under the same runtime monitor with a model that never obeys untrusted content. An agent is relatively secure if every leak it has is a leak the reference already has. Our framework, PlanTERS, instantiates relative security, a notion from speculative execution, with a small agent language in which the attacker chooses how the model follows instructions, a composition with runtime monitors, and attacker models that differ in what they observe. A fence theorem shows that revoking injectable tools at the first untrusted read, and never granting them back, makes every plan relatively secure at the event level on finite runs, for every model behaviour in which each injection is a single call and the model obeys on a schedule fixed in advance; with a fairness condition, this extends to possibly infinite runs. We prove or refute relative security for agents with fences, budgets, approvals, loops and dual LLMs, in Isabelle/HOL.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-09 01:20 · 最新观测版本

On the Relative Security of Prompt Injection: Mechanised Definitions, Defences and Verdicts for LLM Agents

An LLM agent that reads untrusted content, such as an email or a web page, may follow instructions hidden in it. Noninterference is too strong for such agents, since a useful agent lets untrusted content influence what it does and may disclose the user's data by design, and declassification needs a release policy written for each task. We instead state security relative to a reference agent, which runs the same plan under the same runtime monitor with a model that never obeys untrusted content. An agent is relatively secure if every leak it has is a leak the reference already has. Our framework, PlanTERS, instantiates relative security, a notion from speculative execution, with a small agent language in which the attacker chooses how the model follows instructions, a composition with runtime monitors, and attacker models that differ in what they observe. A fence theorem shows that revoking injectable tools at the first untrusted read, and never granting them back, makes every plan relatively secure at the event level on finite runs, for every model behaviour in which each injection is a single call and the model obeys on a schedule fixed in advance; with a fairness condition, this extends to possibly infinite runs. We prove or refute relative security for agents with fences, budgets, approvals, loops and dual LLMs, in Isabelle/HOL.

查看此版本解读 →查看此版本原文 ↗