Google Gemini Adds New Setting to Opt Out of AI Training While Keeping Chat History

Google has rolled out a standalone toggle allowing users to disable Gemini AI model training without sacrificing their conversation logs, according to PCWorld.

Users can now keep the blue toggle for History set to On while separately turning off model training. Previously, Google linked the “help improve our models” toggle with the Gemini app history, forcing users to lose saved conversations if they wanted to opt out of model training.

To update the settings, users can navigate to their Gemini History settings, ensure the history toggle remains on, locate the “Help train generative AI models with” heading, and uncheck options including “Chat content including text & attachments” as well as “Audio and Gemini Live videos & screenshots.”

Google said that the changes are rolling out in the coming weeks. The update addresses privacy concerns regarding how AI providers use chat data for training models, a practice that carries risks of data leakage through malicious prompts involving user credentials, email addresses, private records like legal contracts or medical diagnoses, and other personally identifiable information.

While opting out stops model training, Google updated its policy regarding data retention. Rather than keeping chats for 72 hours, Google may now retain them for “around two months” for “security and safety purposes,” which includes helping to “detect, report, and investigate complex emerging abuse scenarios.”

That retention window is about twice as long as the policies at OpenAI and Anthropic, where both providers state they may keep chats for up to 30 days, though all providers maintain exceptions for security and legal considerations.

Photo of author

Technology Desk

Technology Desk at Archyde.

GTA 6 Meme Token: 95% of First-Hour Volume From Creator’s Own Wallets