Key Take-Aways

• How to architect and secure AI agents: sandboxing, strict permission models, prompt hardening, audit trails
• Threat modelling specific to agent-based systems: prompt injection, data poisoning, malicious plugins etc.
• Offensive security labs: hands-on red teaming for AI agents to understand how attackers exploit weaknesses
• Understanding and using Model Context Protocol (MCP), secure tool orchestration and supply chain in agent frameworks
• Practical code & framework exposure (Python), cloud labs; real exercises for both defense & offense

‍