Building Secure and Responsible LLM Applications Training Course

Overview of LLM Architecture and Attack Surface

• How LLMs are constructed, deployed, and accessed through APIs.
• Essential components of LLM application stacks (e.g., prompts, agents, memory, APIs).
• Identifying where and how security challenges emerge in real-world scenarios.

Prompt Injection and Jailbreak Attacks

• Defining prompt injection and assessing its potential risks.
• Scenarios involving direct and indirect prompt injection.
• Jailbreaking techniques used to circumvent safety filters.
• Strategies for detecting and mitigating such attacks.

Data Leakage and Privacy Risks

• Preventing accidental data exposure through model responses.
• Addressing PII leaks and misuse of model memory.
• Designing privacy-aware prompts and retrieval-augmented generation (RAG) systems.

LLM Output Filtering and Guarding

• Utilizing Guardrails AI for content filtering and validation.
• Establishing output schemas and constraints.
• Monitoring and logging potentially unsafe outputs.

Human-in-the-Loop and Workflow Approaches

• Determining when and where to integrate human oversight.
• Implementing approval queues, scoring thresholds, and fallback mechanisms.
• Calibrating trust and the role of explainability.

Secure LLM App Design Patterns

• Applying least privilege and sandboxing to API calls and agents.
• Implementing rate limiting, throttling, and abuse detection.
• Achieving robust chaining with LangChain and prompt isolation.

Compliance, Logging, and Governance

• Ensuring the auditability of LLM outputs.
• Maintaining traceability and managing prompt/version control.
• Aligning systems with internal security policies and regulatory requirements.

Summary and Next Steps