Protecting Google DeepMind's foundational AI models from unauthorized distillation, which poses a significant threat to intellectual property and integrity.
Requirements
Demonstrated research or product expertise in a field related to adversarial ML, model security, model evaluation, pre-training, post-training, or distillation.
Experience designing and implementing large-scale ML systems and/or counter-abuse infrastructure.
Deep technical expertise in one or more of the following: model distillation, model stealing, Reinforcement Learning, Supervised Fine-Tuning, embeddings analysis, or security.
Strong software engineering skills and experience with ML frameworks like JAX, PyTorch, or TensorFlow.
Responsibilities
Lead the workstream dedicated to protecting Google DeepMind’s most valuable AI assets.
Grow and manage the current team of Research Scientists, Research Engineers, and Software Engineers responsible for both detecting unauthorized distillation and mitigating these threats.
Oversee the full lifecycle of defense from researching novel detection systems and identifying capability theft, to deploying core mitigations and contributing to the model training.
Set the research and engineering roadmap for the Distillation Defense workstream, communicating this to stakeholders and your team.
Lead the effort to set protective policies for Core Model Capabilities.
Partner closely with Product, Legal, and other key teams across GDM and Google to ensure defenses are integrated into the Gemini ecosystem and to influence the broader standard for responsible AI defense.
Other
Ph.D. in Computer Science or a related quantitative field, or a B.S./M.S. in a similar field with 5+ years of relevant industry experience.
2+ years of experience in technical leadership or people management, managing research scientists or software engineers.
Build, manage, and mentor a diverse team of researchers and engineers.
Foster a culture of technical excellence and creative problem-solving in a high-stakes, adversarial environment.
Work with leadership to define acceptable risk levels and trade-offs between model defensibility and performance.
Experience interacting with leadership and cross-functional partners (e.g. Legal, Product).
