ByteDance has built itself an AI safety team. The company’s AI research division, Seed, has set up a dedicated department for safety, alignment, and risk management, and it is now recruiting researchers worldwide.
The unit is called the Model-Alignment & Preparedness department. The announcement went out internally around October 13, 2026. Global recruitment followed soon after.
What Seed is building
The new department will start small. Seed plans to hire an initial group of ten researchers.
The team will get access to resources comparable to those of Seed’s cutting-edge model development teams, with long-term investment planned.
Yan Lin is leading the department. Lin reports directly to Wu Yonghui, the head of Seed.
Open roles are based in three cities: Beijing, Shanghai, and San Jose, California.
The research agenda
The department’s priorities include:
Model behavior specifications. These are the written rules for how a model should act.
Safety evaluations. These are the tests that check whether a model actually follows those rules once it is out in the world.
Automated red-teaming. Red-teaming means deliberately attacking a system to find its weak spots. Automating it means using software, often other AI models, to probe for failures at a scale human testers cannot match.
CBRN preparedness. This covers high-risk capabilities related to chemical, biological, radiological, and nuclear threats.
Why the structure matters
The unit is reportedly the first standalone safety-and-alignment department at a Chinese AI lab. The restructuring places safety functions at the same level as model capability research.
The stated goal is to integrate safety and alignment directly into how Seed develops technology. Prior to this reorganization, efforts related to safety and alignment at Seed and other Chinese AI labs were generally housed within broader Responsible AI sectors, focusing on content moderation and compliance.
The geopolitical backdrop
International conversations about AI governance and risk communication are intensifying, particularly between the US and China. A major Chinese lab creating a dedicated preparedness team, with an explicit focus on threats like CBRN misuse, gives that conversation something concrete to point to.
What this means
For now, Seed has put the structure, the reporting line, and the hiring plan in place. Ten researchers is a modest starting point for an agenda that spans behavior design, evaluations, automated attacks, and CBRN risk. How quickly the team grows will say a lot about how seriously the commitment holds up over time.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.