AGENTIC AI AGENT SECURITY for business

AI agents go beyond chat: they can read documents, use tools, search systems, trigger workflows, interact with APIs, store memory and support business decisions. This course explains the new attack surface and teaches participants how to prevent, detect and control risks such as prompt injection, tool misuse, confidential data leakage, poisoned memory, unsafe automation and human approval traps.

• Foundations: AI assistants vs AI agents; autonomy, tool use, memory and RAG.
• Agentic attack surface: prompts, context, external content, APIs, tools, identity, secrets and human review.
• Critical risks: prompt injection, hidden instructions, data exfiltration, poisoned memory, unsafe automation and reviewer manipulation.
• Secure design: least privilege, allowlists, source ACLs, short-lived credentials, approvals, dry-run mode, logging, kill switches and rollback.
• Prompts and tools: safer system/developer prompts, operating rules, sensitive-data rules, escalation logic and safe output formats.
• Workshop: secure a realistic agent use case, tool permissions, approval steps, release gates and runtime alerts. Risk identification and mitigation.

Instructor-led training with slides, demonstrations, practical examples, case-based discussion, scenario analysis, prompt review, threat modelling, control mapping and a two-hour hands-on workshop using an individual or team-based agent use case.

Instructor-led training with slides, demonstrations, practical examples, case-based discussion, scenario analysis, prompt review, threat modelling, control mapping and a two-hour hands-on workshop using an individual or team-based agent use case.