主题演讲 _ 智能体人工智能：理解和保障智能系统的下一个前沿领域.pdf

1、Agentic AI:Understanding and Securing the Next Frontier of Intelligent SystemsDiana Kelley,CISO,Noma Security Gartner“Intelligent agents in AI are goal-driven software entities.they can receive instructions,create a plan and use tooling to complete tasks,and produce dynamic outputs.”https:/ Differen

2、cesGen AIContent creation based on patterns and promptsSingle model responds to specific requestsPassive response generationLimited to training data boundariesAgentic AITake independent action to achieve specific goalsMultiple specialized agents collaborateProactive problem solvingLeverages external

3、 tools and dataExample Use CaseThe Venice Vacation of your Dreams,Agentic EditionAgent ArchitecturesSingle Agent ArchitectureAgentic AI-Threats and Mitigations,https:/genai.owasp.org/download/45674/Core Architectural ComponentsModel Context Protocol(MCP)The ConnectorOpen standard for resource access

4、 introduced by Anthropic in November 2024Wide adoption,including OpenAI and Google DeepMindAgent to ToolOriginal Image source:modelcontextprotocol.io/introductionAgent2Agent(A2A)The CommunicatorOpen standard for agent interoperability introduced by Google in April 2025Wide industry supportComplement

5、s MCPImage source: source:a2aproject.github.io/A2A/latest/#a2a-and-mcp-complementary-protocolsThreat Modeling-MAESTRO(Multi-Agent Environment,Security,Threat,Risk,and Outcome)https:/cloudsecurityalliance.org/blog/2025/02/06/agentic-ai-threat-modeling-framework-maestroThreat Modeling Steps-MAESTRO Sy

6、stem Decomposition Layer-Specific Threat Modeling Cross-Layer Threat Identification Risk Assessment Mitigation PlanningImplementation&Monitoringhttps:/cloudsecurityalliance.org/blog/2025/02/06/agentic-ai-threat-modeling-framework-maestroThreat Model Overviewhttps:/genai.owasp.org/download/45674/What