Advanced Red Teaming & Adversarial ML

Security considerations for deployed AI systems

Build systems that automatically discover vulnerabilities

Understanding and preventing attacks on safety research

Understanding and defending against prompt injection attacks

Defense mechanisms against adversarial attacks

Attacking AI systems through combined text, image, and audio

Creating and studying controlled examples of misaligned AI behavior

Understanding and preventing data poisoning attacks on AI systems

Advanced techniques for creating adaptive adversarial examples that generalize across models

How LLMs adapt to different programming languages and paradigms