4 ms·
Given that these companies 'own' the software with which their employees are communicating, and 'own' the hardware their employees are using, it should be entir
by danpalmer 9y ago
Given that these companies 'own' the software with which their employees are communicating, and 'own' the hardware their employees are using, it should be entirely possible for them to extract key logs from either or both of those endpoints, which would mean they could happily use TLS 1.3 with no problem.
However, these organisations have a) invest large sums of money in packet-capture appliances that do all this for them on their network, b) deployment of endpoint logging to all client machines is 'hard', and c) they usually don't actually own the software services - they licence ancient things that won't do key export and still have decades until they are EOL'd.
They claim that weakening TLS 1.3 is the only way, but really it's the _cheaper_ way, and they just don't want to take the alternative.
- johncolanduoni 9y agoIt's also much easier to circumvent than a device (which drops anything it cannot understand) physically locked somewhere where the majority of the users cannot access it. Do you have a client-based solution, hard or easy, with new or old software, that provides similar resistance to tampering?
- danpalmer 9y agoImplementing this server-side would probably be the safest option, and would be very tamper resistant. The main problem is just that implementing it on the many pieces of legacy software with expensive support contracts is difficult.
- sametmax 9y agoWell, unless we reach the limit of physics, "we can't" is always synonymous with "we don't want to". And it's usually a matter of resources, values or skills. But it's bad PR to say "we don't want to spend that much" or "we are not skilled enough to do it". So big entities always say it's against moral/law to do it, or that it can't be done.
- oneplane 9y agoOr they could just use their 'enterprise configuration management' to enforce TLS 1.2 internally... They pretty much have no case whatsoever on making TLS 1.3 weak. If TLS 1.3 were to have no enforced FS, they might as well call it TLS 1.2.. oh wait, we already have that.
- voidlogic 9y ago>Or they could just use their 'enterprise configuration management' to enforce TLS 1.2 internally... Until the PCI auditors come and start telling everyone that 1.2 is obsolete because 1.3 exists...
- AaronFriel 9y agoA network setup that permits network admins to obtain plaintext account data (information the PCI spec requires be protected) would not be PCI compliant.
- voidlogic 9y agoI don't think that is true, if the access to decrypt packet captures adheres to PCI access controls why would that violate anything? So long as the decryption of a packet capture happens on host in PCI scope, the user is a PCI user and their actions are directly traceable to that single user? People who run systems in PCI need to debug their production systems too...
- AaronFriel 9y agoThese are all important caveats, but I'd suspect that many/most network devices do not have the rigorous controls to enforce all of those requirements simultaneously. I am out of my element here, though.
- voidlogic 9y agoIts not so bad, you have some dedicated debugging PCI host that you ship your packet captures too (packet capture never leaves PCI), then you checkout necessary private keys (system with keys is in PCI and only lets PCI users do this) and have a debugging session.
- ploxiln 9y agoThe easier way is to install their own CA in all internal-network devices, proxy/MitM all TLS traffic, and effectively block TLS traffic that does not trust/use their internal CA. (Easier way to guarantee all applications are "in compliance".) They can use TLS 1.3 in this model. So yes, agreed, enterprises could deal with TLS 1.3, but they'd rather the rest of the world use worse crypto instead. They'd always rather do nothing to upgrade/fix their systems. They can't help it - it seems less risky and costly, and it's their job to make an attempt to save that risk and cost. So you need to treat them like a natural phenomenon.
- danpalmer 9y agoThat could be easier, but I wouldn't be certain of it. Between BYOD policies and apps certificate pinning it could still be difficult or a non-starter for some applications.
- emmelaich 9y agoYes, certain websites (banks) or specific app endpoints might be whitelisted but for general use I think will give rise to modified browsers. Or versions of ssl libraries will which dump plaintext and/or intermediate for analysis. Either way this is an enormous amount of work for all organisations.
- ploxiln 9y agoI think proxy/MitM is perfect when combined with BYOD and app cert pinning. The user needs to opt-in to the end-to-end-security undermining corporate network. An app can opt-out of such networks. In which case you'll just have to use it at home, or over the cellular data connection. An app that is sympathetic to or sold for "enterprise" use will allow use of the user-added MitM CA in the system root CA store. Finally, a "guest network" could be created if the enterprise wanted to enable users to use BYOD and any app as they are normally used. It's just a matter of everyone making an honest choice between end-to-end security and "enterprise" interception. That was always the choice.
- kevin_nisbet 9y agoSorry, I don't believe your statement to be correct. When using an algorithm with perfect forward secrecy, you negotiate an encryption key for the session with the remote endpoint. The CA's are used just to authenticate the endpoints in the key exchange. This is the reason for the complaint, with forward secrecy, even if you compromise a CA key, you still can't decrypt any traffic.