Applied Scientist - Trust and Safety (Multimodal Foundation Model) - Global Frontier Tech Recruitment Program - 2027 Start (PhD)

Build frontier multimodal trust & safety systems that protect users from harmful content. Work on a multimodal moderation foundation model using large-scale sparse MoE training and cross-modal alignment, and on agentic moderation that uses reinforcement learning for multi-step decision-making with tool ecosystems and interpretable reasoning. Tackle robustness across 200+ languages and adversarial AIGC attacks, and deliver research outcomes with real-world moderation feedback loops.