Is my data used to train AI?
No. There are two parties involved and they are worth separating, because only one of the two answers is a promise we are in a position to make.
Us. We do not use your content to train, fine-tune, distill, or customize any model. We run no model-training pipeline at all and hold no training dataset drawn from your account. That is our own commitment, and it is a term of the Terms of Use (Section 6.3) rather than a description of how things happen to work this month.
The models we call. AI processing runs on our servers, and the models themselves run at Amazon Web Services, through Amazon Bedrock. Under the AWS Service Terms, content sent to Bedrock is not used to train the underlying foundation models and is not shared with the third-party model providers. That is what AWS's own published terms provide, and we tell you so in those words on purpose: AWS's systems are not ours to inspect, so we describe the terms we rely on rather than warrant someone else's conduct. What is ours to answer for is how we have configured it — Section 4.2 of the AI & Data-Training Disclosure names the three Bedrock features that would cause prompt or response content to be kept, states that we have enabled none of them, and describes the automated check that fails our build if that changes.
Your Google data is used to provide the assistant's features — triage, summaries, drafting, and the reasoning behind them — under your own authorization. For the authoritative account of how AI processing works and what is and isn't used for training, see the AI & Data-Training Disclosure and the Privacy Policy.
See also: Where does my data live? · How is my data secured?