SSH is only trust-on-first-connect if you choose to behave that way. Whenever you connect to a new machine, it prompts you "I don't recognize this machine, and its fingerprint is XXXX. Do you trust that?"

I don't see how the option to trust websites is worse than having a bunch of certificate authorities choose who I'll trust for me.

How would you possibly know which fingerprints to trust? Also, do you honestly think there is any hope for the average user to understand what that means and know what to trust and what not to trust? IMO, that is a massive step backwards in usability, which directly impacts the overall security of a given solution. If something is not usable, then people will figure out a way around it and security then goes out the window.

For example, if users were responsible for knowing which fingerprints to trust for a given website, they would most likely just click "ok, trust it" for everything. Then, you're overall security goes way down because now people are conditioned to click "accept" to everything, regardless of the impact.

Signed archives of trusted or untrusted fingerprints, distributed by various and independent authorities is one option.

Trust is, by definition, an extention of solidity or support. CA is a trust model, which has proved both brittle and unworkable.

http://www.etymonline.com/index.php?term=trust&allowed_in_fr...

Google and other services presently provide extended trust and validity assessments for websites: pinned certificates, malware scans, and the like. A limited number of such reliable schemes would scale reasonably well, and should prove useful.

I'm not saying "perfect", I'm saying "useful".

DANE perhaps?

http://www.internetsociety.org/deploy360/resources/dane/

Interesting, that's a possibility.

I was thinking of something more akin to, say Google's pinned certificates (something which practices such as Let's Encrypt actually makes harder AFAICT):

https://security.stackexchange.com/questions/29988/what-is-c...

Or something that might be a parallel of email reputation services -- SenderBase / IronPort (now Cisco) rating email servers by their spam loads, etc.

Rather than negative reputation, a positive reputation (vouch rather than warn) might be viable. (Negative ratings systems, digital or otherwise, tend to inspire various legal assaults.)

As an individual user, you have no better information than your knowledge of the website and its claims on security, maybe its brand, and what your social circles claim about its security practices. A certificate authority at least has the potential to aggregate data and from there assess the security of a website. The original ideas behind certificate authorities are not dissimilar from insurance.