What AI Labs Know About You Through Your Coding Agents and Chats

The author conducted a personal experiment to analyze what AI labs and inference providers can learn from their daily AI usage, discovering that coding agents leak significantly more sensitive data than standard chat interactions. By running a local model on their own hardware to inspect exported transcripts and session logs, the author found that local models could successfully rebuild detailed personal profiles, routines, and professional histories. The author argues that standard privacy settings and chat toggles are insufficient because tools and agents automatically ingest extensive local machine data. Open article →