研究人员开发了一种新颖的黑盒方法来检测人工智能应用程序中护栏系统的存在和功能。该方法通过监控行为信号，包括 HTTP、词汇和时序数据，在事先不知道系统的情况下区分护栏拦截和 LLM 拒绝。实验表明，该方法在检测护栏方面达到了 100% 的准确率，并能识别出它们旨在阻止的具体内容类别，区分 LLM 拒绝的 F1 分数为 98%。 AI
影响 通过阐明护栏行为，能够对人工智能系统进行更有效的对抗性测试和安全审计。
排序理由 学术论文，详细介绍了一种新的人工智能安全研究方法。[lever_c_demoted from research: ic=1 ai=1.0]
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2607.02121v1 Announce Type: cross Abstract: As Large Language Models (LLMs) and agentic systems become integrated into real-world applications, ensuring their safety and security is critical. Guardrail systems that detect and block malicious instructions sent to and from an…
As Large Language Models (LLMs) and agentic systems become integrated into real-world applications, ensuring their safety and security is critical. Guardrail systems that detect and block malicious instructions sent to and from an LLM are an essential component of AI security. Ho…