ByteDance's Seed AI Lab Establishes Model Safety and Alignment Department

ByteDance's Seed AI lab is creating a distinct department focused on model alignment and preparedness. Yan Lin is appointed to lead this new unit, which will report to Seed head Wu Yonghui. The department will absorb members from the existing Responsible AI team and some personnel from the Chat team.

The newly formed department will be responsible for developing model behavior specifications, conducting safety evaluations, implementing alignment training, and establishing protective systems and risk monitoring. Its scope includes ensuring AI agents operate within human intentions and authorized limits, as well as addressing potential chemical, biological, radiological, and nuclear risks associated with advanced AI models.

Seed AI also intends to increase public collaboration with external researchers and independent evaluators. This organizational change creates a dedicated structure for safety and alignment efforts as the lab progresses with developing more capable and autonomous AI systems.