|
所在平台: Udemy |
课程主页: https://www.udemy.com/course/agent-name-service-ans-for-secure-ai-agent-discovery/
课程评论：没有评论
Coursera 课程总结：Agent Name Service (ANS) — 安全 AI 代理发现 本课程深入探讨了用于安全 AI 代理发现的 Agent Name Service (ANS) 框架。课程首先介绍 Agentic AI 和多智能体系统 (MAS)，阐述独立、面向任务的智能体如何在数字生态系统中运作。 接着，课程详细解析了 ANS 的核心架构，包括代理解析器、信任权威和联邦注册中心。重点关注代理注册生命周期，强调如何利用公钥基础设施 (PKI) 和数字证书实现代理安全、可追溯的注册、续订和吊销。 课程还介绍了 ANS-N Ame 格式，这是一种直观的分层命名系统，可在每个代理名称中嵌入身份、能力、版本和合规性信息。通过版本协商、签名验证、TTL 执行和端点验证等机制，确保了代理发现和交互的健壮性和实时性，同时防止冒充和滥用。课程也讨论了命名冲突和域所有权等治理挑战，并与 ICANN 注册系统进行了比较。 “协议适配器层”模块详细介绍了 ANS 如何通过能力卡、元数据模式、基于角色的策略和安全委托框架来支持各种代理交互（A2A、MCP、ACP）。此外，课程还深入讲解了身份建模和验证，包括零知识证明 (ZKP)、JWT、OAuth、双向 TLS 和沙盒强制执行，用于在运行时对代理进行身份验证和隔离。 在高级部分，课程利用 MAESTRO 7 级威胁模型分析了注册表中毒、DoS 和侧信道攻击等安全漏洞，并提出了 ANS 特有的缓解策略。最后，课程评估了集中式与分布式注册表、联邦解析以及混合缓存模型（Redis、Memcached）等实现选项，以安全高效地扩展 ANS。
This course offers a comprehensive foundation in Agent Name Service (ANS) for Secure AI Agent Discovery, focusing on how autonomous agents securely identify, verify, and collaborate through the Agent Name Service (ANS) framework. We begin by establishing a clear understanding of Agentic AI and Multi-Agent Systems (MAS), framing how independent, task-oriented agents function within intelligent digital ecosystems. From there, learners explore the core architecture of ANS, diving into components such as agent resolvers, trust authorities, and federated registries. Special emphasis is placed on the Agent Registration Lifecycle, highlighting how agents are registered, renewed, and revoked in a secure, traceable manner using Public Key Infrastructure (PKI) and digital certificates.The course then examines how agent discovery and interaction are governed through structured semantics, introducing the ANSName format-an intuitive, hierarchical naming system that embeds identity, capability, version, and compliance in each agent name. Key mechanisms such as version negotiation, signature verification, TTL enforcement, and endpoint validation ensure robust, real-time resolution and prevent impersonation or misuse. Students will also learn about governance challenges, including naming collisions and domain ownership, with comparisons to ICANN-style registries.A full module is devoted to the Protocol Adapter Layer, explaining how ANS supports varied agent interactions (A2A, MCP, ACP) through capability cards, metadata schemas, role-based policies, and secure delegation frameworks. This is paired with deep dives into identity modeling and verification, including the use of Zero-Knowledge Proofs (ZKPs), JWTs, OAuth, mutual TLS, and sandbox enforcement to authenticate and isolate agents at runtime.Advanced sessions explore security using the MAESTRO 7-Layer Threat Model, analyzing vulnerabilities like registry poisoning, DoS, and side-channel attacks, and presenting ANS-specific mitigation strategies. Finally, learners evaluate implementation options such as centralized vs. distributed registries, federated resolution, and hybrid caching models (Redis, Memcached) to scale ANS securely and efficiently.
