Agentic Software Security (Online) - Felix Schumacher - German
Prompt injection, tool misuse, and uncontrolled tool interactions create attack vectors when operating and using agent-based systems and LLM-powered applications, leading to consequences such as data exfiltration or unauthorized system access. These are real risks that software architects and developers should understand and address.
In our 2-day training course “Agentic Software Security,” we’ll teach you the fundamentals you need to know.
The training will enable you to understand attack vectors such as direct and indirect prompt injection, insecure tool interactions, supply chain risks (e.g., tool poisoning or rug pulls), and cross-context effects; methodically assess risks during the development and integration of agentic applications; and prevent them in a targeted manner with appropriate security measures.
Agenda
Attack Vectors on Generative AI
- Identify attack vectors on agent-based systems, such as prompt injections
- Systematically identify trust boundaries in typical agent-based architectures using threat modeling
- Secure agent-based systems through guardrails, sandboxing, proper identity management, and fine-grained authorization
- Secure AI systems throughout their entire lifecycle, from onboarding (e.g., via the MCP Registry) to offboarding
Agent-Based Systems and Protocols
- Understand what distinguishes chatbots from copilots, with a focus on security aspects
- Understand and evaluate security mechanisms of new agent-based standards, e.g., MCP or A2A
- Know what matters when implementing agent-based systems, both technically and organizationally
- Use IAM and authorization patterns strategically to design secure agent-based systems
Your Trainers
Felix Schumacher
INNOQ
IT-Security
- Agentic Software Security
- OWASP Top Ten in practice
- Web Security
Felix is a Senior Consultant at INNOQ. He enjoys working with IT security, test-driven development and the operation and further development of existing systems.