· Red Teaming Testing for frontier multimodal/reasoning LLMs and Agents (Llama, ChatGPT, Claude, Gemini, etc.) using the latest adversarial prompting methods from published research, as well as innovating new jailbreak strategies for multimodal generative AI
· Documenting novel jailbreak risks of frontier LLMs and products related to privilege escalation and harmful content, ensuring models meet both company and global standards (e.g. child safety, hate speech, privacy, counterterrorism, data leakage, and more)
· Performing safety assessments of new products involving foundation models according to industry benchmarks keeping stakeholders apprised of model vulnerabilities
· Serving on an accomplished team of AI scientists, software engineers, lawyers, and subject matter experts (linguistics, psychology, technology, etc.) to balance safety and utility of open-source LLMs