독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · MCP协议安全중대2 분 소요DisconfirmAI 편집부

Snyk Engineer Silverman Explains MCP Security: Distinguishing Malicious Tool Descriptions from Code Vulnerabilities, Demonstrating Figma Injection Attack

Micah Silverman of Snyk presented AI Security Engineer Foundations at the AI Engineer conference. Using a Figma MCP server command injection case, he demonstrated how to distinguish malicious tool descriptions from ordinary code vulnerabilities, emphasizing shifting controls from system prompts to architectural constraints like restricted database views.

2026-10-11 21:30 GMT+8

In a video course at the AI Engineer conference, Snyk security engineer Micah Silverman detailed core mechanisms for AI application security. He used a student chatbot as a 'capture the flag' exercise, noting that when the model refuses to reveal grades, the refusal itself exposes its access to that data. Silverman argued that relying solely on system prompts to keep data private is insufficient because the data remains within the model's available context; instead, architectural measures like restricting database views are necessary to remove the model's access to sensitive information entirely.

Regarding the popular Model Context Protocol (MCP), Silverman proposed a specific security analysis methodology. He distinguished between 'malicious tool descriptions' and 'ordinary code vulnerabilities,' recommending pairing agent scans with dependency and source code analysis. To illustrate the practical impact of this distinction, he cited a command injection case in a Figma MCP server. This example showed how attackers could exploit instructions within tool definitions to execute unauthorized operations, a risk distinct from traditional code logic errors that leverages the AI agent's trust boundary regarding tool descriptions.

The course also covered OWASP risks, shadow AI, threat modeling, and red teaming. Through a live EVO demonstration, Silverman showed how to inventory models, datasets, and MCP servers, then add policies to immediately flag prohibited models already present in a project. He extended this inventory into proposed build gates and reviewed how threat modeling must adapt when prompts, retrieved content, and generated outputs cross trust boundaries. For engineers working on agentic development, this methodology offers a concrete path from traditional application security controls to addressing AI-specific risks.

로그인 / 회원가입
독서리서치레이더투자 프레임워크

출처

  • youtube.com↗
주제

MCP协议安全

Model Context Protocol (MCP) 自2024年11月由 Anthropic 推出以来,已成为连接大语言模型与外部工具的事实标准。随着其被捐赠给 Linux Foundation 并更新至 v2 版本,安全风险日益凸显…

7 단계1 건업데이트됨 2026-10-27 19:06

현재까지의 과정

  1. 2024-11-25Anthropic 发布 Model Context Protocol (MCP) v1 初始规范,旨在标准化 AI 应用与外部工具的集成。cheatsheetseries.owasp.org ↗
  2. 2026-01-24Maloyan 和 Namiot 发表研究,首次对 MCP 规范进行形式化安全分析,识别出三类协议级漏洞,并提出 ATTEST MCP 扩展方案。arxiv.org ↗
  3. 2026-04-01Aviral Srivastava 独立报告 MCP Python 和 TypeScript SDK 中的 OAuth 发现机制缺陷,可能导致敏感凭证泄露至攻击者控制的服务器。medium.com ↗

아직 해결되지 않은 문제

  • MCP v2 的无状态设计和强制 OAuth 是否能在实际生产环境中有效遏制工具描述投毒和间接提示注入?
  • Linux Foundation 接管后,社区能否建立统一的 MCP 服务器安全认证或信誉体系,以解决当前‘周末项目’式的快速部署带来的安全隐患?
전체 주제 읽기 →
리서치 계속하기 →