Then I thought about HSTS and how it makes it easier to connect to a site securely. Sure, it has nothing to do with transport or encryption directly, but the aim was to keep the connection secure by preventing an attack on a browser's ability to *not* use SSL. And I remembered that browsers like Chrome ship with implicit lists of sites which should have HSTS enabled by default. And then it hit me.
With an HSTS-enabled flag for a website in your browser, if you also shipped a certificate fingerprint, it basically bypasses the need for a Certificate Authority.
Think of SSH. What's the one time your connection is in peril?
The authenticity of host 'syw4e.info (18.104.22.168)' can't be established. RSA key fingerprint is 57:f9:cf:53:3a:fb:a4:af:e0:96:3c:20:99:30:82:8e. Are you sure you want to continue connecting (yes/no)?
That question is all that stands between you logging into a real server and a fake server, and establishing a secure connection or not. If you had that key fingerprint already you would know if it was authentic, and you could go on with your life without answering stupid questions.
This is exactly what happens when you visit a site with a self-signed certificate, only it's much more complicated than it needs to be. If your browser simply had a list of those fingerprints (similar to the list of HSTS-enabled websites Chrome has) it could connect securely, automatically, without having to verify against a 3rd-party Certificate Authority.
Though this would make the browser's connections about as secure as with SSH, this isn't practical. There are lots of websites out there. We can't possibly keep a list of all of them in the browser. But if you believe in the idea that HSTS makes us more secure than without it, simply accepting and keeping the first certificate you got would be just as secure, right? Well, there's a problem there.
Websites are constantly changing their certificates. They add and remove hardware regularly, and sometimes revoke certificates if there's suspicion their private key might have been compromised. They also change before they expire. So even if we had a list of the initial certificates' fingerprints, what happens when they change?
In order to support a dynamic network of secured systems, there would need to be an extension to the encryption protocol that allowed downloading updates for future certificates. In addition, sites could publish a list of 'trusted' hosts (perhaps even on different domains) which can also update the fingerprint, so that if host A.B.C is down, you can still get an update for it's certificate fingerprint from B.B.C, or even D.C. In this way, a very simple peer-to-peer network of whitelisted hosts could share updates about the security of the network, without being explicitly tied to a few for-profit corporations (SSL CAs).
So then you may ask, what about internal servers? How can we ensure a user going to a closed, internet-less site knows they're connecting securely? The answer there is, of course, hypocrisy.
If you want your client to be connected securely, you have to exchange some secrets. It's mandatory for encryption to be secure. Maybe it's a HMAC'd challenge-response pair, or a shared key or certificate. You need to share something ahead of time.
With SSL, it's been the chain of certificates from Certificate Authorities that live in your computer and in your browser - up to 650 or more of them! So we can keep that system alive, and keep paying CAs for the 'extra assurance' we need for things like online banking, and offline encrypted connections. But we can also share our own secrets.
Take the case of the 'pay-for-wifi' connection abroad. You connect to the wifi and try to check your e-mail, and are redirected to a page asking you to put in your credit card details. Well wait just a minute! Is that the real page, or did some hacker put that up? With CAs you would be assured, because you have their chain of trust in your browser. But if you don't want to use CAs, you could input the certificate fingerprint yourself, perhaps if it was printed on the wall next to the access point's SSID and WPA-PSK passphrase.
But I digress. The main thing to take away from this post is how HSTS has dramatically changed our perception of 'secure web'. Instead of demanding that all connections are secure, we accept that on the very first visit, we might be getting a 'real' HSTS response from a website, or there might be an attacker lying in wait.
Of course i'm not as smart as I seem. Somebody else has already thought of all this and created it, and is trying to get the browsers and big internet players to buy in. It's not going so well. But since HSTS is already implemented in extremely popular browsers, they must have already accepted the idea of the no-assured-trust-on-first-visit model of security - for the HTTP protocol, anyway. If they accept that, then they're only a step away from the same security that SSH depends upon.
Considering all this, it seems that there's a disconnect in the reality of browser security. The browser and big-internet-guys already assume their connection might be compromised on the first visit. Yet they won't accept this new model that avoids the need for Certificate Authorities. Once implemented, SPDY adoption might actually skyrocket, because the protocol wouldn't be beholden to paying third-parties and depending on all 650 of them for security. I just hope progress and the pursuit of better security wins out over commercial interests.