{
  "version": "https://jsonfeed.org/version/1.1",
  "title": "MAREF 博客 — 智能体治理",
  "description": "关于智能体治理、多智能体安全、形式化验证和构建可信 AI 系统的思考。",
  "home_page_url": "https://maref.cc/zh/blog",
  "feed_url": "https://maref.cc/zh/feed.json",
  "language": "zh",
  "expired": false,
  "items": [
    {
      "id": "https://maref.cc/zh/blog/mcp-integration-tutorial",
      "url": "https://maref.cc/zh/blog/mcp-integration-tutorial",
      "title": "如何通过 MCP 把 MAREF 接入你的 Agent —— 分步教程",
      "summary": "Model Context Protocol（MCP）是 Claude Code、Cursor、Windsurf 调用工具的标准协议。本文展示 MAREF 如何嵌入这条链路——同时作为 MCP client 和 MCP server——让每一次工具调用都经过治理层。",
      "date_published": "2026-08-14T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/langgraph-integration-tutorial",
      "url": "https://maref.cc/zh/blog/langgraph-integration-tutorial",
      "title": "LangGraph 集成 MAREF：今天就能用的协议级治理",
      "summary": "LangGraph 是最流行的基于图结构的 Agent 编排框架。本文展示如何通过 MCP 和 A2A 在今天就把 MAREF 治理放在 LangGraph 节点上——并如实说明原生 sidecar 适配器的进展。",
      "date_published": "2026-08-14T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/performance-benchmarks-are-public",
      "url": "https://maref.cc/zh/blog/performance-benchmarks-are-public",
      "title": "MAREF 的性能数字是公开的 —— 我们这样测量",
      "summary": "我们不要求你相信营销图表——MAREF 把 SLA 基准直接放在代码库里。本文列出我们断言的每一项指标、复现它们的准确命令，以及每个数字对生产 Agent 意味着什么。",
      "date_published": "2026-08-14T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/autonomous-agent-evolution",
      "url": "https://maref.cc/zh/blog/autonomous-agent-evolution",
      "title": "自主智能体进化 —— AI 系统的安全递归自改进",
      "summary": "AI 智能体的递归自演进：Lyapunov 验证的收敛、红蓝对抗训练与宪法式安全约束，如何实现无失控风险的自主改进。",
      "date_published": "2026-06-23T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/from-steadicam-to-agent-governance",
      "url": "https://maref.cc/zh/blog/from-steadicam-to-agent-governance",
      "title": "我拍了 30 年电影，现在我在给 AI Agent 造安全系统",
      "summary": "30 年 steadicam 摄影师，从好莱坞片场到 AI 安全基础设施。听起来像转行，但片场教会我的事，恰恰是 AI 行业最缺的：安全不是速度的敌人，安全是你敢加速的前提。",
      "date_published": "2026-06-23T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "Frankie"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/choose-agent-governance-framework",
      "url": "https://maref.cc/zh/blog/choose-agent-governance-framework",
      "title": "如何选择 Agent 治理框架 — 2026 对比",
      "summary": "LangGraph vs CrewAI vs AutoGen vs MAREF：Agent 治理框架采购指南。对比多智能体部署中的运行时安全、形式化验证、状态机与生产就绪度。",
      "date_published": "2026-06-23T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/multi-agent-runtime-safety",
      "url": "https://maref.cc/zh/blog/multi-agent-runtime-safety",
      "title": "为什么多 Agent 系统需要运行时安全保证 — 2026 指南",
      "summary": "多 Agent AI 系统的运行时安全：身份隔离、熔断器、漂移检测与爆炸半径控制。防止 Agent 失控与级联故障的生产级模式。",
      "date_published": "2026-06-23T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/tla-plus-formal-verification",
      "url": "https://maref.cc/zh/blog/tla-plus-formal-verification",
      "title": "不是「测过」，是「证过」——为什么 MAREF 用 TLA+ 形式化验证",
      "summary": "大多数安全工具是「测过」的。MAREF 是「证过」的——使用 TLA+ 模型检验、Lyapunov 稳定性分析和数学收敛性保证。",
      "date_published": "2026-06-16T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/owasp-agentic-top-10",
      "url": "https://maref.cc/zh/blog/owasp-agentic-top-10",
      "title": "OWASP Agentic Top 10：MAREF 如何覆盖全部 10 项关键风险",
      "summary": "OWASP 发布了 Agentic Top 10——多智能体系统的权威威胁模型。MAREF 是首个覆盖全部 10 项风险的开源框架。",
      "date_published": "2026-06-16T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/88-percent-incidents",
      "url": "https://maref.cc/zh/blog/88-percent-incidents",
      "title": "88% 的公司已经翻车了——关于 AI Agent 安全的 7 个数字真相",
      "summary": "Gartner 说 40% 的企业应用将有 AI Agent，但 88% 已经出过事。花在'用 AI 做安全'上的钱是'保护 AI 本身'的 17 倍。仅 6% 的企业有成熟策略。",
      "date_published": "2026-06-13T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/vibe-coding-crisis",
      "url": "https://maref.cc/zh/blog/vibe-coding-crisis",
      "title": "GitHub 正在变成一个巨型 AI 代码垃圾场",
      "summary": "6.3 亿仓库，一半新代码是 AI 写的，信任度从 77% 跌到 60%。CMU 挖出 600 万假 star，AI 代码严重问题是人工的 1.7 倍。",
      "date_published": "2026-06-13T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    },
    {
      "id": "https://maref.cc/zh/blog/why-agent-governance-matters-2026",
      "url": "https://maref.cc/zh/blog/why-agent-governance-matters-2026",
      "title": "为什么智能体治理在 2026 年至关重要",
      "summary": "多智能体系统正在进入生产环境。没有治理，每个智能体都是数据泄露、幻觉传播和对抗性利用的未监控向量。",
      "date_published": "2026-06-13T00:00:00.000Z",
      "date_modified": "2026-06-16T00:00:00.000Z",
      "authors": [
        {
          "name": "MAREF Engineering"
        }
      ]
    }
  ]
}