
Agentic AI Security: OWASP Agentic Top 10, MCP, NIST and A2A
Udemy ยท English
About this course
- Language
- English
- Category
- Provider
- Udemy
- Data updated
AI agents no longer just answer questions. They read your email, query your systems, call tools and submit transactions, often without a human approving each step. The security controls most organizations rely on were built for request-response applications, and they only partly hold when software plans and acts on its own. This course shows you how to secure, monitor and govern AI agents in a real enterprise, with hands-on labs you run yourself. Learn agentic AI security the way it is practised: threat model, defend, detect, respond and govern, using the OWASP Top 10 for Agentic Applications. Threat model an AI agent and score its risks with OWASP AIVSS and a four-level autonomy scaleDefend against goal hijack, indirect prompt injection and human-agent trust exploitationControl tool use, code execution and irreversible actions with an enforceable tool policySecure the Model Context Protocol (MCP) and Agent2Agent (A2A) communicationGovern agents as non-human identities with least privilege and Microsoft Entra Agent IDDetect agent attacks in your SIEM, respond to incidents and map controls to NIST AI RMF, ISO/IEC 42001 and the EU AI ActWhy agentic AI security matters nowThe OWASP Top 10 for Agentic Applications, released in December 2025, gave security teams a shared list of the risks that appear once AI starts taking actions: goal hijack, tool misuse, identity and privilege abuse, supply chain compromise, unexpected code execution, memory poisoning, insecure inter-agent communication, cascading failures, human-agent trust exploitation and rogue agents. Real incidents such as EchoLeak, the LiteLLM package compromise and the Replit production deletion show these risks are not theoretical. At the same time, MCP and A2A are changing quickly, and regulators expect evidence that autonomous systeโฆ











