Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
- Personal Google AI subscriptions use saved Gemini chat history by default to train generative AI models and improve Google services.
- Having a paid subscription does not exempt users from having their conversation history utilized for model training.
- Transcripts may be selected for review by human contractors, with Google explicitly advising users not to submit confidential or sensitive information.
- Chats evaluated by human reviewers are retained for up to three years and are not removed even if a user deletes their Gemini activity.
- The only method to prevent training is disabling "Keep Activity" in settings, which forces Gemini to delete chat sessions after 72 hours.
Hacker News Discussion
- Industry-wide practice: Commenters emphasize that frontier AI labs routinely log interactions, retain flagged sessions for safety moderation, and utilize human reviewers.
- Data flywheel economics: Several participants argue that consumer AI subscription fees are subsidized primarily because user interactions provide high-value training data and real-world workflows.
- Ecosystem data exposure: Users express concern over Gemini's workspace integrations (Gmail, Google Docs, Photos), questioning whether context pulled from personal accounts becomes vulnerable to human review.
- Legal and compliance skepticism: The community debates how "safety" clauses in Terms of Service and ambiguous GDPR interpretations allow providers to retain and analyze interaction data with minimal consumer recourse.