An open research framework that tests whether language models change their sensitive-data judgments when patient context is added, and whether those changes match the shifts physicians show.
Presented early evidence that context-aware LLMs can classify sensitive health data for consent-driven record sharing.