They train on your data. Even when you use the api. You have to both have a GCP account, be using the right model name and the api. “Preview” no “experimental”. This is so convoluted and hidden that there’s an argument to be made that that’s by design.
both OpenAI and Anthropic promise they don't train on user data when opted out and submitted cuts the API. You don't have to believe them, but that's their claim, anyway.
