You are shipping an application with an LLM at its core. The threat model is different. The transformer has no architectural wall between your system prompt, the user's message, and whatever you retrieved from the database. Every token is processed identically. Defenses must be built around the model, not inside it.
Applied Technical Deep Dive
|
|
What you´ll learn |
Prompt injection and jailbreaking —why input sanitisation alone cannot work, and what architectural boundaries actually containthe blast radius
Sensitive data in context —how system prompts leak, why retrieval pipelines exfiltrate by design, and how to structure trust zones across the context window
Insecure output handling —LLM output is untrusted data; how it reaches downstream systems and how to intercept it before it does damage
Supply chain and model integrity —third-party models, plugins, and fine-tuning pipelines as attack surfaces; what to verify andwhen
Detection —what prompt injection, data exfiltration, and abuse look like in your logs, and how to build signals that fire in production
Mapped to OWASP Top 10 for LLM Applications 2025, MITRE ATLAS v5.5, and NIST AI RMF 1.0.
Pick the path that matches your team – Read more
|
|
Target audience |
Senior Engineers, AppSec Leads, Architects
|
|
Prerequisites |
Working C# / .NET. No prior ML experience required.
