TechNode reported on October 13 that ByteDance's Seed AI lab is creating a Model-Alignment & Preparedness department. It cited an internal email reported by The Paper. Yan Lin will lead the unit and answer to Wu Yonghui, who heads Seed. Staff from the Responsible AI team will move in, along with some people from the Chat team. The unit will write rules for how models should behave, test their safety, and train them to follow human aims. It will also build protective systems and monitor risk. Its remit includes keeping AI agents within what people intend and allow, and chemical, biological, radiological and nuclear risks.
TechNode's summary does not give a start date or a headcount. It also does not say how many staff come from each team. It says Seed plans more cooperation with outside researchers and independent evaluators, without further detail.
This is analysis, not sourced fact. Alignment training teaches a model to follow set behavior rules. Safety evaluations are tests that check whether it does. Agents act on their own, so limits on what they may do matter to buyers. Buyers can ask vendors which safety tests they run and which results they will share.