Tianyi Chen

Tianyi is an Applied Research Scientist at ServiceNow AI Research, where she focuses on adversarial benchmarking and defense evaluation for agentic AI systems. Her recent work centers on RedBench, a multi-generational benchmark designed to evaluate AI agent security and resilience. Tianyi’s research interests span AI safety infrastructure, with particular depth in prompt injection vulnerabilities, runtime action monitoring, and co-evolutionary defense systems. She combines theoretical research—studying frameworks like causal attribution for security and logical air-gap architectures—with applied platform work, investigating ServiceNow’s agentic workflows and long-term memory systems. Earlier in her career, Tianyi contributed to multi-lingual LLM benchmarking and evaluation efforts, establishing evaluation methodologies across diverse languages and models. She has also conducted extensive research in deep research agent benchmarking, exploring how to systematically assess documental retrieval, insight filtering and synthesis. Her research has been published at top-tier venues including ICML and ICLR.