ByteDance’s Seed hires for an AI safety team

Seed established an Alignment & Preparedness team and said it would invest over the long term in AI safety for its increasingly capable models.[1] The unit announced a global recruitment drive for 10 researchers.[1] Seed’s recruitment notice is dated October 14, 2026, and Reuters reported the announcement on October 15.[2][1] The roles are based in Beijing, Shanghai and San Jose, California.[1] Seed says the team will receive computing capacity and research resources comparable to those available to frontier-model teams.[1]

Seed says the team will develop model specifications and safety policies, build safety evaluation systems, and improve models’ ability to identify and defend against risks.[2] Its planned training research covers pre-training, mid-training, supervised fine-tuning and reinforcement learning.[2] Seed also says it will use attacker-defender self-play to strengthen model safety over time.[2] Other stated goals include more generalizable safety training, better model judgment and monitorability, and improved guardrail systems.[2] The listed research areas also include evaluations and benchmarks for frontier-model and agent safety, as well as preparedness for future model risks.[2]

Seed says the team will work with external partners to validate and improve model safety, advance independent third-party evaluations, and collaborate with universities and research institutions.[2] The company says it will integrate safety research into model training, evaluation and deployment.[2] Seed also says it will provide the compute needed to sustain research on frontier models.[2] Together, these plans place the team’s stated remit across both internal model development and outside evaluation, rather than limiting it to a single testing stage.[2]

Reuters described Seed’s announcement as a rare public disclosure by a major Chinese AI developer of a dedicated frontier-model safety and alignment effort.[1] Reuters also reported that The Paper called Seed the first Chinese frontier AI lab to establish a standalone safety and alignment department.[1] Reuters said scrutiny of AI-agent safety is growing in China and the United States, particularly as developers give agents autonomy to access the internet.[1] It defined AI alignment as efforts to make advanced systems act in line with human goals and safety requirements, especially during complex, multi-step tasks.[1] The move came amid wider U.S.-China discussions of AI guardrails, safety risks and communication channels for major AI-related incidents.[1]