https://guijiagi.com/tags/%E5%AE%89%E5%85%A8/Recent content in 安全 on 硅基 AGI · 智能体学习与测评Hugozh-cn本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权Tue, 30 Jun 2026 09:20:00 +0800https://guijiagi.com/posts/llm-alignment-constitutional-ai/Tue, 30 Jun 2026 09:20:00 +0800https://guijiagi.com/posts/llm-alignment-constitutional-ai/系统梳理大模型对齐技术的发展脉络，从RLHF到Constitutional AI再到2026年最新的对齐方案https://guijiagi.com/posts/ai-alignment-2026-rlhf-constitutional/Sun, 28 Jun 2026 10:50:00 +0800https://guijiagi.com/posts/ai-alignment-2026-rlhf-constitutional/2026年AI对齐技术全景：RLHF、Constitutional AI、DPO及最新对齐方法的深度解析https://guijiagi.com/posts/prompt-injection-attacks-2026/Sun, 28 Jun 2026 10:20:00 +0800https://guijiagi.com/posts/prompt-injection-attacks-2026/2026年Prompt注入攻击全景：从直接注入到间接注入、多模态注入的最新攻击手法及防御方案https://guijiagi.com/posts/openclaw-security-system/Sat, 27 Jun 2026 20:00:00 +0800https://guijiagi.com/posts/openclaw-security-system/深入解析 OpenClaw 的安全体系设计，从本地优先到五层安全机制，理解龙虾如何保护用户隐私和数据安全。https://guijiagi.com/posts/system-prompt-security/Sat, 27 Jun 2026 15:00:00 +0800https://guijiagi.com/posts/system-prompt-security/系统Prompt安全加固指南https://guijiagi.com/posts/agent-least-privilege/Sat, 27 Jun 2026 15:00:00 +0800https://guijiagi.com/posts/agent-least-privilege/智能体权限最小化原则https://guijiagi.com/posts/ai-agent-ethics-framework/Fri, 26 Jun 2026 14:00:00 +0800https://guijiagi.com/posts/ai-agent-ethics-framework/系统构建 AI 智能体伦理治理框架，从抽象原则到工程实现，覆盖价值观对齐、安全护栏、责任归属和审计追踪全链路。https://guijiagi.com/posts/prompt-injection-defense/Fri, 26 Jun 2026 14:00:00 +0800https://guijiagi.com/posts/prompt-injection-defense/系统讲解 Prompt 注入攻击的原理、分类与防御策略，包含红队测试方法、代码实现和生产级防御架构。https://guijiagi.com/posts/agent-security-checklist/Fri, 26 Jun 2026 14:00:00 +0800https://guijiagi.com/posts/agent-security-checklist/智能体产品上线前的安全检查清单，涵盖提示注入防护、权限控制、数据安全、合规审查等 20 项关键检查https://guijiagi.com/posts/ai-red-teaming-guide/Thu, 25 Jun 2026 12:00:00 +0800https://guijiagi.com/posts/ai-red-teaming-guide/构建完整的 AI 红队测试框架，包括威胁建模、攻击分类、测试用例设计、漏洞评分与报告模板。https://guijiagi.com/posts/ai-content-moderation/Thu, 25 Jun 2026 12:00:00 +0800https://guijiagi.com/posts/ai-content-moderation/设计生产级 AI 内容审核系统架构，覆盖文本、图像、多模态内容，实现多级过滤、实时拦截与人机协同审核。https://guijiagi.com/posts/hermes-security-framework/Thu, 25 Jun 2026 10:00:00 +0800https://guijiagi.com/posts/hermes-security-framework/全面解析 Hermes Agent 的五层安全防线体系，涵盖用户授权、危险命令审批、容器隔离、行为审计与供应链安全https://guijiagi.com/posts/llm-safety-eval-guide/Thu, 25 Jun 2026 10:00:00 +0800https://guijiagi.com/posts/llm-safety-eval-guide/LLM 安全评估全维度指南：毒性检测、偏见评估、越狱攻击测试与对齐评测https://guijiagi.com/posts/llm-security-checklist/Thu, 25 Jun 2026 10:00:00 +0800https://guijiagi.com/posts/llm-security-checklist/覆盖输入安全、输出安全、模型安全、基础设施安全与合规审计的 LLM 生产安全完整检查清单https://guijiagi.com/posts/openclaw-security-safety/Thu, 25 Jun 2026 10:00:00 +0800https://guijiagi.com/posts/openclaw-security-safety/本地部署零云端依赖，数据主权归用户所有。OpenClaw 的五层安全体系详解。https://guijiagi.com/posts/ai-alignment-techniques/Wed, 24 Jun 2026 11:50:00 +0800https://guijiagi.com/posts/ai-alignment-techniques/系统梳理 AI 对齐的技术路径、方法对比和实践选择