SecurityOnlineDo Son
Google separates Gemini chat history from AI training controls
Google has split Gemini chat history settings from AI model training, introducing granular privacy controls that let users save conversations without enabling model training. The update addresses a common user concern by separating the storage of chat history from the right to use that data for training future Gemini models. Google said the new toggles will roll out to global users in the coming weeks.
Under the refreshed architecture, two independent controls exist: a History toggle for saving and organising conversation records, and a separate setting to opt out of data being used for AI training. By default, if a user previously had Keep Activity disabled, both new toggles are off, but users can re-enable History without re‑consenting to model training.
The company also outlines data retention rules: when history is disabled, new conversations disappear from the visible interface, but prompts remain in Google’s systems for around two months to support responses, feedback processing, and service protection. Even with training opt‑out, prompts may still be processed for answer generation and safety filtering, and any user‑provided feedback may be reviewed by humans.
Google notes that such conversations and their metadata can be retained for up to three years after decoupling from accounts. For Google Workspace users, administrators retain central history controls, and turning off history could limit some connected apps.