Google is refuting the viral claim that its Gemini AI “steals” and learns from users’ Gmail messages. A company spokesperson clarified that Gmail content was not used to train Gemini, nor were any user settings changed without notification. Google emphasized that Gmail’s Smart Features have existed for years, and importantly, Gmail emails and attachments are not used to train Gemini’s AI models unless explicitly authorized by the user.
The company explained that Gemini training is segregated from Workspace app data, including Gmail, Docs, and Sheets. While commands given directly to Gemini may be stored and used to improve services (depending on regional settings), emails and attachments in Gmail are not pulled for training purposes unless the user explicitly requests their use, such as asking Gemini to summarize a draft.
Confusion arises from Gmail’s Smart Features and personalization settings like Smart Compose and Smart Reply, which use machine learning but do not feed data into Gemini’s general AI model training. Disabling these features only turns off those conveniences and does not affect Gemini training, as Gmail content is not used for that purpose. Misinterpretations of Gmail’s settings have contributed to the spread of these rumors.
Regarding privacy, Google distinguishes between consumer Gmail accounts and Workspace accounts, where administrators control whether Gemini can access organizational documents. Personal Gmail data is protected under Google’s general privacy policies, and enterprise data isolation measures are in place for Workspace. This separation aligns with legal requirements like GDPR, preventing unauthorized use of sensitive data for AI training.
Users concerned about their data can review settings such as Gemini Apps Activity in their Google Account to control data contribution to Gemini improvements or allow human reviewers. In Gmail, users can manage Smart Features and personalization settings, which affect features but not AI training. Workspace admins should check their Admin Console settings to ensure data handling complies with company policies.
In summary, Google states clearly that Gemini AI is not secretly trained on Gmail content. The rumor conflated existing Gmail features with AI training processes and misunderstood the separation between Gemini user prompts and Workspace data, which Google keeps isolated.
Here is a refined version of the article with improved word flow:
Google is addressing a viral claim that its Gemini AI “steals” and learns from users’ Gmail messages. The company insists this rumor is false, clarifying that Gmail content is not used for training Gemini models and no settings were changed without users’ knowledge.
Google explained that Gmail’s Smart Features, which have been available for years, are unrelated to training Gemini AI. The AI model’s training data is strictly separated from what’s stored in Gmail, Docs, and Sheets within Google Workspace. While commands given to Gemini may be stored to improve the service, user emails and attachments are not fed into Gemini’s training unless explicitly requested by the user, such as asking for draft summaries.
Some confusion stems from Gmail’s Smart Features like Smart Compose and Smart Reply, which use machine learning for personalization but do not contribute to Gemini’s general AI model training. Disabling these features turns off their conveniences but does not impact Gemini training, as Gmail data is not used for that purpose.
Google underscores the difference between personal Gmail data and Google Workspace data, where admins control AI access to organizational documents. This policy respects privacy laws such as the GDPR and ensures enterprise data isolation. Personal Gmail accounts follow Google’s overarching privacy policies, which separately govern Gemini Apps Activity and product personalization.
To manage data shared with Gemini, users can adjust Gemini Apps Activity settings in their Google Account and control whether their prompts help improve the AI or allow human reviewers. Gmail users can also toggle Smart Features and personalization settings to manage conveniences but not AI training. Workspace administrators should verify settings to maintain compliance with company policies.
In brief, Google assures that the content of Gmail is not secretly used to train Gemini AI. The rumor conflated existing Gmail features with AI training, overlooking the clear separation of Gemini prompts and Workspace data.
