Chinese technology giant ByteDance is strengthening its artificial intelligence safety efforts by creating a dedicated research team and recruiting 10 researchers worldwide. The initiative comes as concerns grow over the risks posed by increasingly capable AI models and autonomous AI agents.
The new Alignment & Preparedness team has been established within Seed, ByteDance’s AI research unit. The company plans to invest in research that makes advanced AI systems safer and more reliable.
Why ByteDance is building an AI safety team
AI systems are becoming more capable of handling complex tasks with less human supervision. However, giving AI agents greater independence can introduce new risks, particularly when they can access the internet or interact with external systems.
AI alignment focuses on ensuring that AI systems follow human intentions and safety requirements. ByteDance’s new team will work to strengthen these safeguards throughout the development and deployment of its models.
What the new team will work on
ByteDance plans to recruit 10 researchers for the team, with positions available in Beijing, Shanghai and San Jose, California.
The researchers will work on safety training, automated vulnerability testing and tools to monitor and understand AI model behaviour. Their responsibilities will also include testing AI agents in controlled environments and preparing for emerging risks as models become more advanced.
The company says the team will receive computing power and research resources comparable to those available to its advanced AI model development teams.
ByteDance expands its AI safety research
The initiative reflects growing attention to AI safety as technology companies develop more powerful models and autonomous agents. ByteDance also plans to work with universities, research institutions and independent evaluators to improve safety testing.
The company intends to integrate safety research across model training, evaluation and deployment rather than treating it as a final check before release. This approach could help identify weaknesses earlier in the development process.
As competition in advanced AI intensifies, investments in safety and alignment are becoming an important part of developing systems that can be deployed responsibly.
Also read: Google launches Gemini 4 Argon AI model.
