The current default physics-of-trust are dangerous-by-default.
They require you to place open-ended trust in an entity now and irrevocably into the future.
You do this when consenting to a permission dialog, or choosing which silo to use.
Even if you understand the technical implications of your decision (highly unlikely), it's still structurally impossible to make that decision.
The actions of an entity can make them lose your trust.
I trusted 23andMe with my genome 10 years ago… but don't trust them now!
That case was well-known enough that I took the effort to remove my data.
But how many other cases are below our notice, and happening constantly?