LLM Application Security and Red Teaming Specialist

Course focused on threat modeling for LLM applications: attack surface mapping, identifying risks via OWASP LLM Top 10, and documenting threat models. Emphasizes data flows, trust boundaries and tool permission considerations.

Map the attack surface of an LLM app: core concepts and entry points to assess exploitable vectors.

Analyze risks using the OWASP LLM Top 10 (2025): understand prioritized vulnerabilities and their implications for LLM applications.

Review data flows, define trust boundaries and tool permissions; formalize and document the resulting threat model.

This module covers prompt injection attacks and output handling: direct and indirect injections, system prompt leakage, insecure outputs and prompt isolation using policy layers. Focused on concepts and practical mitigations.

Examine direct and indirect prompt injections: attack vectors, examples and detection approaches.

Assess risks from system prompt leakage and its effects on model behavior.

Handle insecure outputs and implement prompt isolation and policy layers to limit unauthorized execution or disclosure.

Module focused on data and supply chain security for RAG and LLMs: RAG poisoning, malicious documents, embedding/vector weaknesses, sensitive data disclosure and risks from dependencies, models and dataset supply chains.

Identify and mitigate RAG poisoning and malicious documents introduced into knowledge stores.

Examine embedding and vector weaknesses and their impact on retrieval and result relevance.

Address sensitive data disclosure and supply chain risks: dependencies, external models and third-party datasets.

Secure LLM agents by enforcing least-privilege controls, curating tool allowlists, scopes and sandboxes, mitigating unbounded consumption and cost abuse, and producing security regression tests and reports tailored to LLM systems.

Analysis of excessive agency risks and applying least-privilege to LLM agents.

Design and governance of tool allowlists, scopes, approval workflows and sandboxes.

Detecting and mitigating unbounded consumption and cost-abuse scenarios in LLM pipelines.

How to design and run security regression tests and produce actionable reports for LLM systems.