By Eduardo Baptista
BEIJING, Oct 15 (Reuters) – Chinese tech giant ByteDance’s AI research unit Seed has created an Alignment & Preparedness team and launched a global drive to recruit 10 researchers, saying it will make a long-term investment in AI safety for its increasingly capable models.
The hiring push comes as scrutiny of AI-agent safety grows in China and the US, especially as developers give them autonomy to access the Internet.
AI alignment refers to efforts to ensure advanced AI systems act in line with human goals and safety requirements, particularly when carrying out complex, multi-step tasks.
ByteDance’s announcement is a rare public disclosure by a major Chinese AI developer of a dedicated effort focused on frontier-model safety, alignment and preparedness for emerging risks.
State-backed newspaper The Paper reported on Tuesday that Seed was the first Chinese frontier AI lab to establish a standalone safety and alignment department.
The move also comes amid broader US-China discussions on AI guardrails, safety risks and channels for communicating about major AI-related incidents, even as the two countries compete for leadership in advanced AI.
As model capabilities improve, safety and alignment are no longer simply safeguards against risk but core capabilities that will determine whether models can be deployed at scale, Seed said in a recruitment post published on its official WeChat account on Wednesday.
Seed said it would provide the new team with computing capacity and research resources comparable to those available to frontier-model teams. The positions are based in Beijing, Shanghai and San Jose, California.
Its research will include safety training across model-development stages, automated testing for vulnerabilities, tools to monitor and explain model behaviour, AI-agent testing environments and preparations for emerging risks.
Seed said it would also seek independent testing and work with universities and research institutions, applying safety research across model training, evaluation and deployment.
(Reporting by Eduardo BaptistaEditing by Mark Potter)
Comments