Outline

Identify the AI threat surface: poisoning attacks, backdoors, prompt injection, and data exfiltration. • Implement defense-in-depth strategies and map purple-team tactics to detections and controls. • Hands-on: Develop policy-driven two-pass RAG with local LLMs, demonstrate robustness using ART/TextAttack, and perform telemetry anomaly scoring.

Outline

Design and implement an LLM misuse detector with a Streamlit mini-dashboard. • Apply provenance verification (batch CLI) integrated with policy actions. • Harden RAG intake with hash/MIME checks, denylists, and canaries.

Outline

Assess the robustness of AI models against adversarial attacks. • Implement and evaluate anomaly scoring techniques for telemetry data. • Compare autoencoder performance against Isolation Forest for outlier detection (optional).

Outline

Define and track Key Performance Indicators (KPIs) for AI security. • Establish effective monitoring strategies for AI systems. • Develop robust incident response plans and conduct After Action Reviews (AAR).

Outline

Map advanced attack techniques to defensive controls and detections. • Practice purple-teaming exercises to enhance AI system resilience. • Implement next-generation protective measures against evolving AI threats.