📄 中文摘要
随着对大型语言模型（LLMs）在关键应用中依赖的增加，系统提示的暴露问题逐渐显现。这一现象源于生成模型的固有限制以及对提示级安全性的错误假设，给专有逻辑、数据完整性和用户信任带来了重大风险。分析揭示了系统提示暴露的机制，强调了基于提示的安全措施的脆弱性，并突显了建立强大技术保障的迫切需求。
通过创意查询揭示 AI 系统内部逻辑：提出增强的访问限制
📄 English Summary
AI System's Internal Logic Exposed via Creative Querying: Enhanced Access Restrictions Proposed
The increasing reliance on Large Language Models (LLMs) in critical applications has revealed a significant vulnerability: the exposure of system prompts through creative user querying. This issue arises from the inherent limitations of generative models and the flawed assumption of prompt-level security, posing substantial risks to proprietary logic, data integrity, and user trust. The analysis dissects the mechanisms behind system prompt exposure, highlights the fragility of prompt-based security measures, and underscores the urgent need for robust technical safeguards.