ByteDance Seed has recently announced the establishment of the "Seed Model-Alignment & Preparedness" department, led by Yan Lin, who reports to Wu Yonghui, the head of Seed. Members of the original Responsible AI team and some Chat teams will join the new department, further strengthening the investment in model safety and alignment.

The new department will focus on enhancing the security capabilities of intelligent agent models (Agent Models) and preventing potential risks posed by emerging capabilities such as chemical, biological, radiological, and nuclear (CBRN) threats. Its work will cover the development of model regulations, the construction of a security evaluation system, alignment training, the establishment of protection systems, and risk monitoring, while also strengthening open collaboration with external research institutions and third-party evaluation organizations.

It is reported that as the capabilities and autonomy of large models continue to grow, security issues such as model behavior deviating from expectations, misuse of agents, and cyberattacks have received more attention. Previously, the safety and alignment work of domestic large model teams was mainly integrated into the overall corporate security compliance system. This adjustment indicates that Seed is establishing safety and alignment as an independent function. According to relevant information, ByteDance Seed is the first frontier AI laboratory in China to establish a dedicated department for safety and alignment.

Seed internal believes that safety and alignment not only relate to risk prevention but also affect whether models can achieve large-scale application in complex scenarios. The new department will strive to ensure that cutting-edge models effectively perform tasks under safe standards and authorized constraints, continuously reduce the risks of model misuse and behavioral deviation, and improve the security system against unknown risks.