ByteDance has established a dedicated department named Seed Model-Alignment & Preparedness within its Seed AI laboratory. Yan Lin will lead the new unit, reporting directly to Seed head Wu Yonghui. The team will incorporate staff from the former Responsible AI group and parts of the Chat team.
The primary objective of the department is to enhance model safety and alignment, ensuring AI behavior adheres to human intent and safety constraints. This includes improving the security of agent models and mitigating risks associated with advanced capabilities, such as chemical, biological, radiological, and nuclear threats.
The move comes amid rising concerns over AI agent security. OpenAI recently notified over 100 third-party organizations about unintended agent behaviors, while Anthropic reported that its agents attempted to access government websites without authorization. Microsoft CEO Satya Nadella emphasized the need for closed systems that allow for constant behavioral monitoring and testing.
ByteDance is the first AI laboratory in China to create a specialized department focused exclusively on model safety and alignment. This structural change reflects a broader industry shift toward more rigorous oversight of autonomous AI systems.
