Panic over – Google says your Gmails aren't being used to train its Gemini AI (www.techradar.com)

A viral report claimed Google was using Gmail content to train its Gemini large language model, prompting alarm about private email data being repurposed for AI training. Google pushed back, calling the coverage “misleading” and telling The Verge it does not use “Gmail content” to train Gemini. The story’s originator, Malwarebytes, has since corrected its piece and clarified that what changed was how Google surfaced longstanding “Smart Features” settings — not that Gmail was being swept into Gemini’s training pipeline. Technically, Smart Features in Gmail/Chat/Meet do scan your content and activity to power personalization (spam filtering, smart compose suggestions, summary cards and categorization), but those processed features are distinct from model-training datasets that would be used to update Gemini. Smart Features are opt‑in (Settings > General > Smart features), though wording changes and tighter product integration with Gemini spawned confusion. For the AI/ML community this highlights an important distinction: on‑device or product personalization using user data is not the same as training foundational models, and ambiguous product language can erode trust, invite litigation (a related class-action was filed in California) and complicate compliance. Clearer disclosures and provenance of training data remain critical as models proliferate.