"Confer" LLM claims E2EE privacy

de0u

"Signal creator Moxie Marlinspike wants to do for AI what he did for messaging" (Dan Goodin, Ars Technica, 2026-01-13)

meowijuana

de0u what lumo could've been, this is pretty cool though.

Probably9857

de0u

I think Maple.ai already does this...

https://trymaple.ai/

de0u

There is also Apple's "Private Cloud Compute".

Remote enclaves can be attacked (see, for example, SGX.Fail). Part of my model is that if a sufficiently well-resourced entity (criminal gang, government intelligence agency, etc.) wants my data in particular that will likely happen. But I'd prefer for it to be hard for the malware gangs to exfiltrate my data, and really hard for them to get everybody's all at once.

ryrona

Attesting the virtual machines running the AI models in this way is a really trust inspiring move, especially if it really is possible to verify the software running in detail, ie if the VM images are provided for download to compare with what is attested.

But nothing of this is end-to-end encryption. It is still classic client-server encryption as is used by any website, the server is still capable of decrypting all your communication. All you did was verify what hardware, software and configuration said server is running. In fact, they need the data in decrypted state to be able to run it in AI models.

This is abuse of terminology, and false marketing, claiming properties of their solutions they do not have.

In the meantime, local AI is actually private.

just4frun

The solution to this is a fully homomorphic encryption. It has to be solved yet but in the recent years a significant progress was made and a lot of discoveries are actually production ready, I even saw an AI startup based on FHE

Pocketstar

I'd rather have my own AI in my own possesion, I use chatterUI on my phone, and I can also link it to my miniPC as well for a slight speedboost, even though it makes you dependant on an active internet connection. (I would switch between online API and offline modes if there is a blackout)
I hardly use it though, but it can be handy as a tool, but be wary of the many hallucinations LLM's and other AI's make, it is best practice to check everything manually, wether you use a commercial tool, or an autonomous one, after all; it is a tool to assist people, not to replace them, the quality is too poor for that.

https://github.com/Vali-98/ChatterUI

Byku

Looks like they only support G****e and 1Password for handling passkey. No way to try it.

fid03

Byku Looks like they only support G****e and 1Password for handling passkey. No way to try it.

Works fine with Proton Pass in Vanadium as well.

(Not going to comment on the service as a whole).