跳到正文
arXiv cs.AI· Ana Clara Zoppi Serpa, Anderson Rocha, H\'elio Pedrini·· 4 小时前AI 评分60

研究:363 个模型中有 296 个被智能体套出系统提示词中的密钥

Eight in Ten LLMs Leak System-Prompt Secrets: Nevan, an Agentic Capture-the-Flag Study

AI 导读

研究团队对 OpenRouter 目录中 363 个文本输入输出模型做指令层级审计,由 OpenCode(基于 DeepSeek V4.1 Flash)在人类研究员引导下诱导模型输出系统提示词中的 16 字符 flag,最终从 296 个模型及 4 个目录冻结后发布的模型中成功提取,共 300 次捕获,其中 170 次借助非英语语言切换、最常用法语。

正文 · 原文

View PDF

Abstract:We audit instruction-hierarchy enforcement across 363 text-in/text-out models from the OpenRouter catalog. The research question we answer with this experiment is which models disclose a protected secret embedded in the system prompt, and through which techniques. Each model guards a secret 16-character flag set in its system prompt, while an artificial intelligence (AI) agent (OpenCode, powered by DeepSeek V4.1 Flash), steered by a human researcher, tries to persuade it to emit the flag. The flag was exfiltrated from 296 of the 363 models, plus four models released after the initial catalog freeze, for 300 captures in total. Of these, 170 involved a non-English language pivot, most often French. Each capture shows that a model ranked a user request above its system instruction at least once. It does not show that the model provably resists untested attacks, nor that it would consistently leak semantically sensitive data, nor that it is vulnerable to harmful-content jailbreaks. For practitioners, we draw three lessons: do not store secrets in system prompts, enforce critical behavior outside the model with output validation and deterministic checks, and do not let a large language model (LLM) judge be the sole decider of consequential outcomes.
Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.21956 [cs.CR]
  (or arXiv:2611.21956v1 [cs.CR] for this version)
  https://doi.org/10.48550/arXiv.2611.21956

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Ana Clara Zoppi Serpa [view email]
[v1] Mon, 16 Nov 2026 22:37:31 UTC (101 KB)

来源:arXiv cs.AI · arxiv.org