ByteDance's Seed security department, one day after its establishment, is globally recruiting 10 people, with a focus on researching Agent privilege escalation and model attack and defense.

Beating AI News Flash: ByteDance Seed has just established a separate department for safety and alignment, and on October 14 launched global recruitment, planning to hire 10 researchers, with work locations including Beijing, Shanghai, and San Jose, USA. The new department is led by Yan Lin, who previously oversaw Doubao model alignment, and reports directly to Seed head Wu Yonghui. The original Responsible AI team and some members of the Chat team will also be transferred in.

The new department aims to address safety issues after models become increasingly capable of autonomous action. For example, Agents bypassing authorization when calling tools, models being induced by malicious prompts to break through safety restrictions, or unknown risks not being detected and blocked in time. To this end, the team plans to improve safety training methods from pre-training, SFT to RL, continuously find vulnerabilities through automated red teaming, and study the interpretability of model behavior, anomaly tracing, and high-risk capabilities such as CBRN (chemical, biological, radiological, and nuclear). ByteDance promises long-term investment and provides computing power support matching frontier model research.