AI 安全漏洞：为何保护提示符还不够.pptx

1、AI Security:,Why Protecting Prompts is Not Enough,Welcome and Introductions,Rex Thexton,Senior Managing Director,Rehan Jalil,CEO,Note:Can we add an image which implies total disruptionPerhaps we can ask GPT to create a colorful image of disruption/explosion,AI is disrupting how SW is built&how work

2、is done,4,Unleashing the Power for AI in the Enterprise Requires Safe Use of Proprietary Data with AI,Security for GenAI-Data Controls is#1 CIO Priority to Prepare for GenAI,Tools,Agents,App Services,Typical AI System,Agents,A2A,MCP,Most Interesting AI Security&Trust Frameworks,OWASP Top 10 for LLMs

3、,Gartner AI TRiSM,Infrastructure and Stack,AI Governance,AI Runtime Inspection&Enforcement,AI TRiSM Technology Functions,AI TechnologyTraditional focused on AITraditional Technology,Source Gartner,OWASP Top 10 for LLMs,(Credit:OWASP Top 10 for LLMs),9,Tools,Agents,App Services,Mapping OWASP Top 10 f

4、or LLMs on an AI System,Agents,A2A,LLM01Prompt Injection,LLM01Prompt Injection,MCP,Tools,Agents,App Services,Mapping OWASP Top 10 for LLMs on an AI System,Agents,A2A,LLM08Excessive Agency,LLM08Excessive Agency,LLM08Excessive Agency,LLM08Excessive Agency,MCP,Tools,Agents,App Services,Mapping OWASP To

5、p 10 for LLMs on an AI System,Agents,A2A,LLM06Sensitive Information Disclosure,LLM06Sensitive Information Disclosure,LLM06Sensitive Information Disclosure,MCP,Tools,Agents,App Services,Mapping OWASP Top 10 for LLMs on an AI System,Agents,A2A,LLM02Insecure OutputHandling,LLM02Insecure OutputHandling,

6、LLM03Training Data Poisoning,MCP,Tools,Agents,App Services,Mapping OWASP Top 10 for LLMs on an AI System,Agents,A2A,LLM04Model Denial of Service,LLM04Model Denial of Service,LLM04Model Denial of Service,LLM05Supply Chain,MCP,Tools,Agents,App Services,Mapping OWASP Top 10 for LLMs on an AI System,Age