• Adversarial Machine Learning: Attacking target machine learning pipelines, manipulating output decisions via input data poisoning, executing model evasion, and reconstructing training secrets using model inversion.
• Securing Large Language Models (LLMs): Defending generative AI applications against direct/indirect prompt injection, bypassing safety constraints (jailbreaking), and preventing training-set data leakage.
• AI for Defensive Security Operations: Leveraging machine learning algorithms to identify anomalous behavioral baselines, automate responses via Security Orchestration, Automation, and Response (SOAR), and generate synthetic data for testing.