ByteDance Seed on October 13 establishes new first-tier unit Seed Model-Alignment & Preparedness to strengthen CBRN and agent model safety

ByteDance's Seed team announced via internal email on October 13 the establishment of the new first-tier Seed Model-Alignment & Preparedness unit, formed from the former Responsible AI team and parts of the Chat team. Yan Lin leads the unit and reports directly to Seed head Wu Yonghui. This is the first time ByteDance has elevated large model alignment to an independent first-tier department. The unit's first priority is strengthening safety controls for agent models, and second is preventing CBRN frontier risks covering chemical, biological, radiological, and nuclear categories. Responsibilities include model specification, safety evaluation, and alignment training, aimed at building safety guardrails for products with autonomous task-execution capabilities. The reorganization comes amid rapid iteration of agent products in the broader industry and accumulating concerns about uncontrolled autonomous agents, reflecting that leading companies are placing agent safety governance ahead of product release.