You know there are more polite ways to ask me for sources.
I suppose I should start by saying this is starting in the US, and not all of these features are necessarily available(or available for users) everywhere.
Photos has an "Opt-in" that has Gemini scan all photos in your album and generate images of you and your friends/family: https://www.forbes.com/sites/zakdoffman/2026/04/20/google-starts-scanning-all-your-photos-as-new-update-goes-live/
Photos employs AI moderators that scan your photos and remove anything they deem violates their policies
Photos has an AI search feature
So they have an open admission that you can opt-in to having AI train on your photos, as well as two features that are passively looking at them. Of those two the search feature in particular is open about the fact that they do train on your prompts. In order to do that they also need to train off of the photos that you get the results from, which they don't say. Photos is next to inarguable.
Docs is a bit less clear
Google claims that it isn't training its models on them, but I have several reasons to doubt them.
They added AI reading of documents, so you can listen to them.
Multiple people have reported that google has locked them out of their documents, with the only explanation that we can find being the content of the documents. https://freedom.press/digisec/blog/google-docs-locks-out-writer/
Google has already added content filters to docs in the past.
Google has show a pattern of behavior, that ignores user will and privacy, a short look at their history of fines over privacy law violations will show this. Hundreds of millions of dollars in individual fines, and yet despite this, they haven't stopped.
These two things stand
Google used to have a slogan, "Don't be evil"
Statement number one is only true in the past tense.
Even assuming they aren't actively training their models on everything right now, they have already begun implementing training on user data, and all signs indicate this will only increase. It takes a single update for an "opt-in" feature to become "opt-out" or even non-optional, assuming Google even honors the "opt-in" which they have proven before they are willing to violate(see below). The largest AI models already have everything easily available, and they are beginning to look for new material. Google is sitting on top of one of the largest sources of human writing, including material that no other AI model has access to.
Do you really trust them not to?
I wish you luck, but I don't think Google is too concerned about the law, or the potential penalties they would face.
That's over a billion dollars just from ignoring privacy laws in the last couple of years. I didn't even list all of them, or their other legal fines for violating laws unrelated to privacy. In total Google was fined almost 4 Billion dollars in legal penalties in 2025 alone. https://proton.me/tech-fines-tracker