ChatGPT thumbs up may override training opt-out

The author examines how providing feedback such as thumbs up or down on consumer AI chat apps can override user privacy and training opt-out settings, causing conversations to be stored and used for model training or human review. Different platforms like ChatGPT, Gemini, Claude, and Copilot handle rated conversations and data retention in varying ways, ranging from multi-year storage to exclusions for foundation models. Additionally, the article warns that work accounts generally allow administrators and employers to access or export employee chat histories. Open article →