The longer dream of a universally federated, decentralized trust mesh depends on a substrate protocol emerging for which I think Bitcoin is only the first glimmer of hope; timeframe ca.10-50 years.
The longer dream of a universally federated, decentralized trust mesh depends on a substrate protocol emerging for which I think Bitcoin is only the first glimmer of hope; timeframe ca.10-50 years.
I'd much sooner support adoption of Marlinspike's Convergence or something similarly distributed: http://convergence.io/
Where it fails is as a protection against state sponsored scrutiny or a serious direct hacking attempt.
It too has seen not a lot of traction. However, Google has implemented a competing stack into Chrome (and Firefox supports it) in HPKP (HTTP Public Key Pinning) with sites having the ability to pin their cert, and these browsers will trust that bond all day.
Might as well place the CA authorities with the TLDs, then. At least that would limit their authority to their subtree, instead of the free-for-all that is the current CA club. It might even let end-users make a semi informed trust decision: ".be seems to run a tight shop and appears trustworthy, .ly not so much" (random TLDs picked for example purposes only).
It would even make it worthwhile for site owners to pick a more reliable/valuable/expensive TLD (and thus CA), since it would bring a generally-accepted-to-be higher level of trust.
Trust is hard. For top-down TLS CA to work, the net needs a fewer trustworthy, sensible, well-secured/geo ip'd/leadership nonprofits to be the systems-of-record far out of reach of govt and corporate dictatorships. (icann, iana are a good start, but they could be better. ssl cert cert authority as another type common-good org.).
Also, having a zillion CAs able to independently issue endless wildcard certs for the same domain is lunacy... only the domain owner can be issued certs. It should be tied to common global record-keeping system as domain name registrars do and/or domain registrar-mediated challenge to prove ownership (control panel login / authenticated API calls, not just adding some spoofable TXT record cookie or link in an email). End MITM by rogue certs, at a minimum.
Heck, make it a requirement that registrars issue free SSL certificates as part of the domain registation service and we might even get to the https-everywhere world that seems to be in vogue. Double bonus if this can be engineered with a technical guarantee/verification that only the proper registrar for a domain is a valid CA for it.
DANE actually reduces the people you need to trust. Hell, some CAs will issue a cert for a domain to anyone who can write to the docroot on the webserver it points to. Grabbing one before a malicious entity and deploying HSTS and HPKP is your only weak defence.
I wrote about this recently if you're interested. http://metafarce.com/adventures-in-dane/
"..the purpose of DANE is not to do away with the CA system. It is to provide another chain of trust based on the DNS hierarchy."
The commercial CAs will be a relic. It's only a matter of time before the whole house of cards collapses. The price pressure has been downwards for years. EV is not meaningful to most end-users, remaining a money-printing scam that provides little assurance of safety (if Enron, or Ashley Madison could get one, what's it worth?), and is vulnerable to MITM and rebinding attacks[1].
I anticipate that Amazon's CA[2] could undermine the profitability - and via desperation, the integrity - of the whole sector even further.
[1] http://www.blackhat.com/presentations/bh-usa-09/ZUSMAN/BHUSA... [2] http://www.geekwire.com/2015/amazon-wants-to-be-your-ssl-cer...
For extra sneakyness, they could probably return the regular glue records and whois information to most requestors, and poisoned values to the targeted CA, to get the certificate up an running in advance of hijacking the domain in general.
The problem is that you can't mitigate insecure CAs by adding a new set of irrevocable government-controlled CAs.
That's the point 'idlewords is making. Right now, the DNS does not function as a sort of distributed CA. But that's exactly what DNSSEC asks it to do.
Unless we're talking about removing domain validated certificates, I don't see why explicitly treating DNS as a CA, when it's a de facto CA is a problem. I'm not sure how we can get to HTTPS everywhere without domain validated certificates?