Chloe Li

I spend my time thinking about how to make powerful AI systems honest and aligned with human values. I’m currently a member of technical staff at Anthropic, where I work on alignment training.

Previously, I worked on AI safety research like model spec midtraining for aligning models, honesty training, chain-of-thought monitoring and control evaluations. I also spent time on AI safety field-building, like leading ARENA (a ML engineering program for upskilling people in technical AI safety) and directing Cambridge AI Safety Hub, where I helped founded research programs like MARS.

Publications & other work

• Model Spec Midtraining: Improving How Alignment Training Generalizes
  Chloe Li, Nevan Wichers, Sara Price, Samuel Marks, Jon Kutasov (2026)
  

• Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives
  Chloe Li, Mary Phuong, Daniel Tan (2025)
  ICLR 2026

• LLMs Can Covertly Sandbag on Capability Evaluations Against Chain-of-Thought Monitoring
  Chloe Li, Mary Phuong, Noah Y. Siegel (2025)
  IJCNLP-AACL 2025, Oral Presentation

• ARENA LLM Evaluations Curriculum