📄 中文摘要
安全部署大规模语言模型（LLM）不仅仅是将其置于防火墙后。生产环境面临传统安全框架无法解决的攻击向量，包括提示注入、通过上下文窗口的数据外泄、嵌入反演以及代理目标劫持等风险。为应对这些挑战，提供了 12 项可操作的安全实践，映射到 OWASP 的 LLM 应用程序前 10 名（2025）和 Agentic 前 10 名（2026）。每项实践都包含实施代码、威胁背景和优先级指导，帮助企业有效保护其 LLM 部署。
企业 AI 安全：生产环境中部署 LLM 的 12 个最佳实践
📄 English Summary
Enterprise AI Security: 12 Best Practices for Deploying LLMs in Production
Deploying large language models (LLMs) securely in production goes beyond simply placing them behind a firewall. Production environments face attack vectors that traditional security frameworks do not address, such as prompt injection, data exfiltration through context windows, embedding inversion, and agent goal hijacking. To tackle these challenges, 12 actionable security practices are outlined, mapped to OWASP's Top 10 for LLM Applications (2025) and Agentic's Top 10 (2026). Each practice includes implementation code, threat context, and prioritization guidance, assisting enterprises in effectively securing their LLM deployments.