Gmail Now Uses Your Emails to Train Its AI (Even Your Attachments)

Google has quietly expanded Gmail’s AI capabilities by allowing the service to read users’ emails and attachments to train its Gemini models. This data collection is enabled by default, meaning most users are unknowingly contributing personal and corporate information to improve predictive text, Smart Compose, and upcoming inbox‑management tools. The move leverages the massive volume of Gmail content, but the lack of a clear, upfront opt‑in prompt has sparked criticism over transparency and user consent.

For business professionals, the privacy implications are significant. While Google promises anonymization and secure handling during model training, the default setting effectively grants the company access to potentially sensitive communications. Users must manually turn off two distinct settings—Smart Features in Gmail Chat & Meet and Smart Features across Google Workspace—to fully opt out. Disabling these options also disables convenient AI‑driven features, creating a trade‑off between productivity gains and data protection that each organization must evaluate.

The Gmail case reflects a broader industry trend: AI providers are mining existing data troves to accelerate model development, often with minimal user awareness. Companies beyond Google face similar scrutiny as regulators and privacy advocates demand clearer consent mechanisms. Enterprises should audit their email platforms, educate staff on opt‑out procedures, and consider the reputational risk of inadvertent data sharing. Transparent communication and robust privacy policies will become essential differentiators as AI continues to reshape everyday workflows.