Two COLM Papers Focus on Privacy and Agents

tianshi_li · x · 2026-07-09

The author's lab had 2 papers accepted to COLM.

One explores how to build datasets with nuanced human contextual privacy preferences and analyzes why large language models struggle to adapt to individualized privacy preferences. The other investigates the impact of agent autonomy on privacy concerns and trust, finding that users feel a greater sense of control when the system only asks the agent to hand over control upon detecting risks, even surpassing the approach of requiring user confirmation at every step.

Related event: Two COLM Papers Focus on LLM Privacy and Agent Trust(3 posts)→

Original post →

More from Safety

Safety channel →