Would have to be with a very noticable warning. Other than that, it could use TOFU like Gemini.
@Hyolobrika
It almost works like that already.
When you open a page on a server with self-signed cert, it gives you a warning, if you accept it, it adds an exception for that cert — you can see the list in preferences under Privacy & Security → Certificates → View certificates → Servers
@Hyolobrika
It also keeps the fingerprints so if you get a different cert on a later visit, it will give you a warning again.
To simplify adding an exception on the first visit you might want to consider this: http://kb.mozillazine.org/Browser.xul.error_pages.expert_bad_cert
You don't need permission from a certificate authority then, much more independent.
@Hyolobrika @Hyolobrika
Self-signed certs do not provide the capability to revoke them. Imagine that a malicious actor isn't just spoofing the site you trust with their own self-signed cert, but that the private key got compromised. With self-signed certs you have no way of telling users that the already trusted certificate is no longer valid, such a capability implies some sort of infrastructure and infrastructure implies hierarchy as someone has to operate it🤷
Why can't TLS do the same thing?
But yeah theoretically you could have a private CRL server if we could get OSes and browsers to let us configure it
@feld @Hyolobrika
Another problem with HTTPS and self-signed certs is that while Firefox was created in the days when self-signed certificates weren't considered bad. However, relying on centralised authorities became a tradition since, so newer software might be designed differently, e.g. rustls which has become the dominant Rust crate for TLS stuff, doesn't allow handling TOFU and self-signed certs cleanly: https://github.com/rustls/rustls/issues/435 for them it's "dangerous_configuration" 😩
@Hyolobrika @feld
Just noticed this reply, sorry. No, it's not just about stating the intent clearly, it's more of a build flag — you have to build this crate with support for handling such cases, by default this code just gets excluded, and as this isn't a default feature, I suppose it doesn't get tested as thoroughly, it might take time for the devs to notice that something is broken in that code. Which IMO is a quite different approach.
