🇨🇳 ByteDance Creates AI Safety Unit

ByteDance’s AI research team Seed has set up an independent model safety and alignment department focused on improving AI agent security and reducing risks from advanced models in chemical, biological, radiological and nuclear domains. The new unit, called Seed Model-Alignment & Preparedness, will be led by Yan Lin and report directly to Seed head Wu Yonghui.

The move consolidates ByteDance’s existing Responsible AI team and part of its chat-model staff into a dedicated structure. Chinese large-model developers have generally handled safety and alignment within broader compliance systems; the report said Seed is the first frontier AI lab in China to establish a standalone department for the function.

The new department will cover behavior standards, safety evaluation systems, alignment training, safeguards, and risk monitoring, while expanding cooperation with external research and third-party evaluators. The change reflects growing industry concern over increasingly autonomous AI agents and their potential misuse or unauthorized actions.

#China

@asianomics