AI Safety Researcher

You will conduct frontier AI safety research across large models, AI agents, and multi-agent systems. You will develop benchmarks, investigate red-team attacks and defenses, build quantitative risk models, validate mitigation approaches, and translate research into evaluation platforms, safety services, and governance materials.

Responsibilities

• Design and build capability and safety benchmarks for large models, AI agents, and multi-agent systems
• Curate evaluation datasets and develop automated testing pipelines
• Conduct systematic evaluations and produce technical reports or academic papers
• Research red-teaming attack and defense algorithms
• Identify vulnerabilities and safety risks in AI systems
• Build quantitative risk models for cybersecurity, biological, chemical, harmful-manipulation, and loss-of-control risks
• Conduct risk assessments using evaluation data, expert forecasts, and incident case studies
• Research alignment, monitoring, interpretability, and agent safety methods
• Prototype and validate risk-mitigation solutions
• Integrate research outcomes into risk-monitoring and safety-evaluation services
• Support safety governance reports, standards recommendations, and policy briefs

Requirements

• Bachelor’s degree or above
• Scientific reasoning and complex research problem-solving skills
• Data structures and algorithms
• Python for algorithm design and data analysis
• Knowledge of large-model and AI-agent architectures
• Experience with large-model deployment, API usage, and harness development
• Experience in large-model or agent safety evaluation and risk analysis
• Technical writing
• Chinese reading, writing, and spoken communication