Hacker News Reader

Comment by GuB-42 | Hacker News Reader

Another one that worries me is Let's Encrypt.

It is not as bad as Cloudflare or AWS because certificates will not expire the instant there is an outage, but considers that:

- It serves about 2/3 of all websites

- TLS is becoming more and more critical over time. If certificates fail, the web may as well be down

- Certificate lifetimes are becoming shorter and shorter, now 90 days, but Let's Encrypt is now considering 6 days, with 47 days being planned as a minimum

- An outage is one thing, but should a compromise happen, that would be even more catastrophic

Let's Encrypt is a good guy now, but remember that Google used to be a good guy in the 2000s too!

View on HN

(Disclaimer: I am tech lead of Let's Encrypt software engineering)

I'm also concerned about LE being a single point of failure for the internet! I really wish there were other free and open CAs out there. Our goal is to encrypt the web, not to perpetuate ourselves.

That said, I'm not sure the line of reasoning here really holds up? There's a big difference between this three-hour outage and the multi-day outage that would be necessary to prevent certificate renewal, even with 6-day certs. And there's an even bigger difference between this sort of network disruption and the kind of compromise that would be necessary to take LE out permanently.

So while yes, I share your fear about the internet-wide impact of total Let's Encrypt collapse, I don't think that these situations are particularly analogous.

Agree, I’ve thought about this one too. The history of SSL/TLS certs is pretty hacky anyway in my opinion. The main problem they are solving really should have been solved at the network layer with ubiquitous IPsec and key distribution via DNS since most users just blindly trust whatever root CAs ship with their browser or OS, and the ecosystem has been full of implementation and operational issues.

Let’s Encrypt is great at making the existing system less painful, and there are a few alternatives like ZeroSSL, but all of this automation is basically a pile of workarounds on top of a fundamentally inappropriate design.

There's not really a way around the initial trust problem with consumer oriented certs though. Yours could reduce the number of initially trusted down to one I think but not any further.

its a shame DANE never took off. If we actually got around to running a trusted DNSSEC based DNS system and allowed clients to create certificates thanks to DANE, we would be in a far more resilient setup compared to what we are now.

But DNSSEC was hard according to some, and now we are running a massive SPOF in terms of TLS certificates.

It didn't "not take off" --- it didn't work. You couldn't run it on the actual Internet with actual users, at least not without having a fallback path that attackers could trigger that meant DANE was really just yet another CA, only this one you can't detect misbehavior or kill it when it does misbehave.

I'll bite: DNSSEC isn't a massive SPOF but TLS certificates are?

Google was always a for-profit operation. Let's Encrypt/ISRG could still go rotten but there are less incentives for them to do so as a non-profit.

Reply on news.ycombinator.com