Why are you changing the topic? Let me simplify this.

You wrote:

Sure lets outsource trust in our communications to some shitty machine learning algorithm that is dumber than a fucking toddler.

I disagree this is bad, and my point is: "A dumb machine learning algorithm is enough to be helpful for this purpose of scam warnings."

Of course you shouldn't trust Google in not stealing your data with their implementation, but you have already established that in your original comment, and I didn't bring this up at all because I already agree with you there. What I disagree with is that this is a bad idea. I think trying to protect people from scams is a great use of AI and would love an open source implementation of it that could be included in GrapheneOS for example.