ByteDance's Seed establishes a standalone AI security department: Yan Lin reports directly to Wu Yonghui, with a focus on preventing Agent privilege escalation.

Beating AI News Flash: ByteDance Seed announces the establishment of an independent Safety and Alignment department (Model-Alignment & Preparedness), led by Yan Lin, head of alignment for the Doubao large model, reporting directly to Seed head Wu Yonghui. The original Responsible AI team and some members of the Chat team will be transferred to the new department.

The new department's focus is to ensure that models continue to comply with safety and authorization requirements as their capabilities grow. In addition to preventing chat responses from overstepping boundaries, it must also ensure that Agents do not act without authorization when calling tools or executing tasks, and guard against the misuse of high-risk capabilities in the CBRN (chemical, biological, radiological, and nuclear) domain. The team will be responsible for formulating model specifications, conducting safety evaluations and alignment training, establishing protection and risk monitoring mechanisms, and bringing in external research institutions and third-party evaluations.