安全研究人员在多个模型上下文协议 (MCP) 服务器中发现了重大漏洞，包括来自 Atlassian、GitHub、Cloudflare 和 Microsoft 的服务器。最常见的关键漏洞是间接提示注入，攻击者可以操纵 MCP 服务器获取的数据来诱骗 AI 代理执行恶意指令。其他问题包括通过错误标记的工具权限进行的权限提升以及 HTTP 调用工具中的服务器端请求伪造 (SSRF) 漏洞。这些发现突显了 MCP 生态系统中存在重大的安全风险，近 30% 的扫描软件包存在高或关键严重程度的漏洞。 AI
影响 凸显了 AI 代理集成中的关键安全风险，可能因信任问题而减缓企业采用。
排序理由 安全研究论文，详细介绍了多个 MCP 服务器中的漏洞。
AI 生成摘要 · Google Gemini · 来自 7 个来源。 我们如何撰写摘要 →
<p>MCPSafe (mcpsafe.io) runs automated security scans of Model Context Protocol (MCP) server repositories using a five-model LLM judge panel and a purpose-built scoring rubric called AIVSS (AI Vulnerability Severity Score). Over the past three months, we've scanned 50+ MCP server…
<p>MCP servers are not browser extensions. When you install one, you are adding a process to your system that may have direct access to your filesystem, network stack, environment variables, and shell. It can read files, make outbound HTTP requests, and execute commands — all on …
... and central IT decided to disable the Atlassian agents in Copilot... :/ For Jira, I was already finding the agent very limited in the data it had access to. I'm probably better off exporting Jira data as spreadsheets and handing them over to Copilot to analyze. For Confluence…
Experimenting with the JIRA agent. How it's going: # ai # copilot
Looks like the Jira agent is working as well... # copilot # ai
I'm hoping this will encourage more library apps and databases to provide MCP connectors. # ai # copilot https:// techcommunity.microsoft.com/bl og/microsoft365copilotblog/federated-copilot-connectors---bringing-real-time-enterprise-data-within-microso/4515993