Healthcare can’t govern the AI tools it hasn’t found yet

Paubox survey results indicate that healthcare organizations frequently expose real patient data to AI coding assistants during development and testing, with 93% reporting at least some data reaches the assistants and 70% finding AI-built tools handling patient data outside IT visibility. Many organizations lack written AI coding policies, and respondents report AI-generated code contributing to hardcoded credentials and insecure storage. The report recommends enforcing synthetic test data and tool-based controls because training alone has not prevented patient data from reaching vendor assistants.