4 ms·

While security and privacy are not parallel, they are not orthogonal either. You can have both. Integrity checks can be done anonymously (e.g.: you could have

While security and privacy are not parallel, they are not orthogonal either. You can have both. Integrity checks can be done anonymously (e.g.: you could have a p2p network of devices sharing a signed database of certificate revocations). Encrypting something gives me privacy. Signing something gives me security. Encrypting a signed package gives me both.

dcow 6y ago
They _are_ orthogonal, you’re just saying that you can have some of both which is exactly my my point about them existing on a spectrum. And it’s not as simple as encrypting data. You have to trust somebody to determine what good integrity looks like and to then verify the integrity information is fresh. The same privacy concern exists if you run OCSP against cypher-text as it does plaintext. You still have a stream of all the things people do. Bad for privacy. Running a decentralized system means you have to trust all the nodes to not to store data or collude. Same problem in a different way. You simply cannot achieve integrity verification if you don't trust anyone to do it. And this is why they are orthogonal. Trust is not compatible with doubt. The issue here is that Apple took off the shelf OCSP and applied it in a way it was not designed for. So there _are_ actual problems with their late implementation. They should be fixed. And personally I think OCSP is kinda dumb because it mechanically defeats the advantage of certs (you don’t need a cert if you’re going to phone home for every invocation, just check a hash), but meh.
- noobermin 6y ago
  You're using the extreme cases of each to argue they are "orthogonal". Perhaps at some extreme no one actually lives in, they are orthogonal. Most people don't need extreme privacy or extreme security, so in the cases that matter, this observation (which seems to be a major crux in your argument) is not important.
  - dcow 6y ago
    The point is that the more private you make something the less ability you have to audit its integrity. If sending a 3rd party a list of hashes is a privacy problem, then security is what takes a hit in order to preserve privacy. That’s not an “extreme case”, it’s what’s being discussed in the essay and in this thread. Similar examples include DNS over TLS vs DNS filtering for content security, and client certs for mutual TLS vs exposing personal information in said cert, and secure neighbor discovery, and IPv6 (can’t have a global IP because someone might track it), the list goes on. I’m not saying we should pursue security at all costs or privacy at all costs, far from it. I am saying exactly that there’s a balance between the two and moreover that the balancing point may be different for individual people which leads to arguments like we’re seeing here between people who calibrate more on the security end vs people who prefer extreme privacy. And in my experience people very often conflate the two, which makes it hard to have a productive discussion. Finally, I’d venture to say that the privacy push of late is having impacts on the ability to deploy strong identity because much of the privacy wave lacks the nuance to distinguish between entities you trust and hence with which it’s okay to maintain a stable secure identity, and those that aren’t. Instead the trend lately has been remove stable identifiers (e.g. Apple’s move to fake mac addresses and GDPR’s IPs are PII) and conceal everything no matter what (TLS 1.3 and DoT/DoH although props to Mozilla for making it possible to configure at the network level via DNs).

dcow 6y ago

They _are_ orthogonal, you’re just saying that you can have some of both which is exactly my my point about them existing on a spectrum. And it’s not as simple as encrypting data. You have to trust somebody to determine what good integrity looks like and to then verify the integrity information is fresh. The same privacy concern exists if you run OCSP against cypher-text as it does plaintext. You still have a stream of all the things people do. Bad for privacy. Running a decentralized system means you have to trust all the nodes to not to store data or collude. Same problem in a different way. You simply cannot achieve integrity verification if you don't trust anyone to do it. And this is why they are orthogonal. Trust is not compatible with doubt. The issue here is that Apple took off the shelf OCSP and applied it in a way it was not designed for. So there _are_ actual problems with their late implementation. They should be fixed. And personally I think OCSP is kinda dumb because it mechanically defeats the advantage of certs (you don’t need a cert if you’re going to phone home for every invocation, just check a hash), but meh.

noobermin 6y ago
You're using the extreme cases of each to argue they are "orthogonal". Perhaps at some extreme no one actually lives in, they are orthogonal. Most people don't need extreme privacy or extreme security, so in the cases that matter, this observation (which seems to be a major crux in your argument) is not important.
- dcow 6y ago
  The point is that the more private you make something the less ability you have to audit its integrity. If sending a 3rd party a list of hashes is a privacy problem, then security is what takes a hit in order to preserve privacy. That’s not an “extreme case”, it’s what’s being discussed in the essay and in this thread. Similar examples include DNS over TLS vs DNS filtering for content security, and client certs for mutual TLS vs exposing personal information in said cert, and secure neighbor discovery, and IPv6 (can’t have a global IP because someone might track it), the list goes on. I’m not saying we should pursue security at all costs or privacy at all costs, far from it. I am saying exactly that there’s a balance between the two and moreover that the balancing point may be different for individual people which leads to arguments like we’re seeing here between people who calibrate more on the security end vs people who prefer extreme privacy. And in my experience people very often conflate the two, which makes it hard to have a productive discussion. Finally, I’d venture to say that the privacy push of late is having impacts on the ability to deploy strong identity because much of the privacy wave lacks the nuance to distinguish between entities you trust and hence with which it’s okay to maintain a stable secure identity, and those that aren’t. Instead the trend lately has been remove stable identifiers (e.g. Apple’s move to fake mac addresses and GDPR’s IPs are PII) and conceal everything no matter what (TLS 1.3 and DoT/DoH although props to Mozilla for making it possible to configure at the network level via DNs).