Article

Google splits Gemini privacy toggle: Chats can be excluded from training but may still be reviewed by humans

Beating AI Flash News: Google announced adjustments to Gemini's privacy settings. Users can finally retain their chat history while refusing to allow their conversations to be used for training AI models. Previously, these two functions were tied to the same toggle; to prevent Google from using chat data to train models, users had to give up their history as well.

The new settings separate chat history from model training. Users can independently control whether to save chats and whether to permit Google to use chat text, uploaded files, images, and other content to train models. After disabling training authorization, future conversations will, in principle, not be used for generative AI model training, except when users actively submit feedback.

Google has also extended the data retention period after turning off chat history, increasing it from the previous maximum of 72 hours to about 2 months. The official explanation is that this data is still needed to process user requests, feedback, and security issues.

Even if conversations are prohibited from being used for model training, Google may still have human reviewers examine portions of chat content for security audits and service improvements. Records reviewed by humans will be separated from user accounts and retained for up to 3 years. Even if users delete their chat history, this portion of records may not necessarily be erased.

The new settings will be gradually rolled out to global users over the next few weeks. Users who previously turned off "Keep activity records" will find both the chat history and model training toggles remain off after migration.