← 研踪 PaperTrailarXiv:2611.21956 · v1
Eight in Ten LLMs Leak System-Prompt Secrets: Nevan, an Agentic Capture-the-Flag Study
首次提交 2026-11-17 06:37
版本更新 2026-11-17 06:37
雷达观测 2026-11-18 19:43
摘要
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.
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 · 2026-11-17 06:37 · 最新观测版本
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.
查看此版本解读 →查看此版本原文 ↗