Papers

363 个模型中 296 个泄露系统提示词密钥:Nevan 智能体 CTF 研究

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

Why we picked it

把 363 个模型的系统提示词当靶子逐个攻破,八成沦陷,还发现换成法语问更容易得手。做智能体或 API 集成的开发者该看看那三条防线建议,比单篇测评扎实得多。

Nevan 研究对 OpenRouter 目录中的 363 个文本模型做了指令层级审计:每个模型的系统提示词里放一个 16 字符的 flag,由 OpenCode 驱动的 DeepSeek V4.1 Flash 智能体在人工研究员引导下尝试套取。结果 296 个模型的 flag 被成功窃取,加上目录冻结后发布的 4 个模型,共计 300 次捕获。

Body

攻击路径上有 170 次涉及非英语语言切换,其中法语最常见——说明多语言提示往往绕过指令层级的英文训练偏向。每次捕获证明该模型至少一次把用户请求排在了系统指令之上,但论文明确指出这不能证明模型对未测试的攻击有抵抗力,也不能等同于会泄露语义敏感数据或被越狱出有害内容。

作者给从业者三条建议:不要把密钥存进系统提示词、用输出校验和确定性检查在模型外部强制关键行为、不要让 LLM 当裁判独自决定重要结果。对部署智能体的团队来说,这是对"提示词即安全边界"假设的一次系统性证伪。

Original · arXiv: DeepSeek

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

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.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include DeepSeek, LLM security, system prompts, agents.

Primary source verification: published directly by arXiv: DeepSeek with first-party authority.