📄 中文摘要
随着自主大型语言模型（LLMs）对云计算和操作技术（OT）的直接控制，它们成为拥有机器速度访问API、数据和控制系统的特权内部人员。非人类身份（NHI）的数量将超过人类80倍，使每个代理都成为易受劫持、克隆和提示注入的高价值账户。缺乏运行时防御代理的监控、评分和干预，单一被攻陷的工作流程可能在几分钟内从篡改的遥测数据转变为工厂停机。因此，建立有效的运行时防御机制显得尤为重要。
运行时防御代理：部署防御性人工智能以追踪、遏制和回滚云中的恶意大型语言模型
📄 English Summary
Runtime Defense Agents Deploying Defensive Ai To Hunt Contain And Roll Back Rogue Llms Across Cloud
As agentic large language models (LLMs) gain direct control over cloud and operational technology (OT), they become privileged insiders with machine-speed access to APIs, data, and control systems. Non-human identities (NHIs) will outnumber humans by 80 to 1, turning each agent into a high-value account vulnerable to hijacking, cloning, and prompt injection. Without runtime defense agents that monitor, score, and intervene, a single compromised workflow can quickly pivot from tampered telemetry to plant downtime within minutes. Therefore, establishing effective runtime defense mechanisms is crucial.