We don't log, but it's just a promise from us and we aren't currently proving it. There are architectures could allow us to prove such, but currently we are just trying to keep up all the new AI stuff.

To argue that our current model is no different than ChatGPT in terms of practical privacy is very wrong though.

If law enforcement comes to us the most we can give them is user agent strings of our users. That is much different than ChatGPT which can provide credit card details, email addresses, etc to link chats with actual people.