Okay I looked over their stuff, a couple thoughts:
I want them to be more clear in their privacy policy about what exactly they can and would reveal for a court order, what their screening process is for those orders, under what conditions they would fight one and if they will reveal anything outside the context of a full court order.
Reason: this is one of your biggest areas of vulnerability when signing up for a phone plan.
The lexipol leaks showed that many police departments use phone information requests so much that they include a set of request forms (typically one for each carrier) in the appendix of their operations manuals. Frequently the forms are the only data request tool in that appendix.
If you happened to have a call with someone who then did something Cool™ and got picked up, expect the detective to have your name and address on a post-it on their desk by the next morning. If you talked to them on some online chat platform they’ll send a court order to that platform for your IP then do the same to your carrier to unmask your identity.
Yes, if you were also sufficiently Cool™ they’ll start doing more invasive things like directly tracking your phone via tower dumps, but that’s a significant escalation in time and effort. If things got Cool™ enough that this is a concern though, it may buy you time to get a new phone if you live in an area dense enough for that to not be immediately identifying.
Also: I suspect the zip code is completely unverifiable so put whatever you want in there, basically pick your favorite sales tax rate.
The LLMs are just somewhere between an averaging and a lossy compression of everything on GitHub. There’s nothing about the current paradigm of “AI” that is going to somehow do better than just rehashing that training set but with the inclusion of various classes of errors.
I think it’s better to view it as spicy search rather than any form of intelligence.