I think it was Schneier who described encryption algorithms as being like a single fence post which is a thousand miles high. The vulnerability is not the encryption algorithm. The attacker is not going to break the encryption, the attacker is going to get access to the key.
The ways attackers might do that have very little to do with cryptography. Espionage, corruption, social engineering, bureaucratic incompetence, completely unrelated vulnerabilities in government networks, etc.
People are repulsed when you call it a back door or a "golden key," but "secure golden keys" are simply key escrows. Escrows are one of the first things you learn to build after asymmetric crypto.
Escrows are well understood, secure, and generally regarded as a good idea. However, "back doors" are scary, sordid, and insecure?
The encryption is using the usual cryptographic primitives. The problem is that the encryption is not the problem.
> Escrows are well understood, secure, and generally regarded as a good idea.
They are not. Because they're the same thing as back doors.
Using key escrow as a method of e.g. password recovery is a security vulnerability. And at least then you're choosing who you trust and the trusted party doesn't have a nuclear neon target painted on it because the keys in escrow are only for specific parties rather than everybody everywhere.
That's really the main issue. It is manifestly unwise to create a system which, if broken, yields the keys to everything. Because it makes the ROI of breaking it so enormous that the quantity and strength of the attackers you attract will overwhelm any imperfect system, and all systems are imperfect.
I have no idea what you mean by "security vulnerability" in this case. Key escrows are the primary way asymmetric crypto is used in practice. An escrow recovers the symmetric key a message was encrypted with and you proceed as normal.
> the trusted party doesn't have a nuclear neon target painted on it
By that logic, government agencies should operate exclusively by typewriter. They don't because crypto primitives are designed to be secure against unrealistically strong adversaries. No crypto paper has ever said, "Our construction is secure in the standard model as long as the ROI is less than epsilon."
You can go down the path of "what if there's another Snowden, and it's a snow day in Kentucky...," but you aren't doing cryptography anymore--you're just developing mild paranoia.
You're using an unusually broad definition of what "key escrow" means.
http://en.wikipedia.org/wiki/Key_escrow
And your definition is covering the scenarios that nobody would call a back door, so I'm not sure what point you're trying to make.
Giving a third party access to your private information is a security vulnerability unless you can trust the third party, and "you can trust the government" is a statement contrary to evidence.
> By that logic, government agencies should operate exclusively by typewriter. They don't because crypto primitives are designed to be secure against unrealistically strong adversaries. No crypto paper has ever said, "Our construction is secure in the standard model as long as the ROI is less than epsilon."
Once again, the crypto is not the issue. Nobody is expecting the attacker to solve the discrete logarithm problem. But breaking the cryptosystem is not the only way to obtain the encryption keys. You have a serious security vulnerability if the expected value of the attacker obtaining the secret key is larger than the amount of money required to bribe the relevant government employee(s). Or if the servers the government keeps the keys on are vulnerable to heartbleed/shellshock/whatever at any point after attackers learn of the vulnerability, regardless of the strength of the escrow cryptosystem.
> You can go down the path of "what if there's another Snowden, and it's a snow day in Kentucky...," but you aren't doing cryptography anymore--you're just developing mild paranoia.
It's not doing cryptography at all. It isn't a cryptographic problem. The strength of the cryptography makes little difference because the cryptography is not the weakest link.
The problem is that government agencies have demonstrated very clearly that they will use such a key as often and as liberally as they can get away with. Every time the key is brought out and used, it becomes vulnerable to interception or disclosure to the point where pretending that it is sure to be safe is just completely out of touch with reality.
I think that's the main point when people disagree about this kind of thing. One side says "but theoretically our fort knox vault solution would be secure" and the other side says "What I just heard you say is 'If I sprinkle magic pixie dust around this intentional vulnerability and ignore it then it would be secure' and that's ridiculous, a group of humans would not almost certainly not act securely around that vault because they never have".
Of course the "other side" is right, but people don't think that way naturally so this kind of proposal/argument keeps coming up.
Key escrow is a back door.
Schneier has generally moved focus from designing and breaking cryptography (although he was a key contributor to the Skein SHA3 candidate) to the bigger aspects of risk. And trusting the government or any third party is risk.
You'll like his blog.
Only the NSA could generate the key initially. After that, who knows who will get access to it? You still have the same fundamental problem as with all backdoors, you've just eliminated one that most of them have (the fact that they're crap enough that you don't even need the back-door key to access them).
"The NSA" isn't a person. If "the NSA" has access to a backdoor that means there are possibly hundreds of people who have access to that backdoor, every single one being a mere human being.
It's not a secure golden key in a glass box with the label "in case of emergency break glass". It's a note on a shared whiteboard in an open plan office.
Predicting the generator's output reduces to solving the discrete logarithm on elliptic curves, which is incredibly hard unless you chose the private key.
Just like in any asymmetric crypto scheme, given only the public key, it's very difficult to find the private key.
(Actually doing it requires knowledge of how the generator works. I promise you it is easy given the private key, but you can go into its internals if you don't believe me.)
> Isn't it just trivial for someone to "recover" your data by pretent to be you by court order, NSL or other means?
Without a "golden key," somebody would have to ask you. With the "golden key," somebody would have to ask the organization in charge OR you.
Edward Snowden proved how this is far from secure. There are people from the NSA, who if they wanted to spy on a girlfriend, could walk out with those keys.
The underlying problem is that you're no longer relying on technological security and mathematical certainty, but rather on human security and the capriciousness of people who earn $65,000/year to keep secrets.
There's no security at all there.
Your assumption that ""golden keys" can be implemented securely" is wrong. The only secure golden keys are those not used. Which means, in order for such keys to be really secure they mustn't be created.
Isn't is true that today, I can easily clone iPhone's data by declaring I loss my phone and need to buy a new one and recover everything from iCloud?
Isn't it just trivial for someone to "recover" your data by pretent to be you by court order, NSL or other means?
Steve's thought on how safe the data in cloud storage is by asking a question: "How easy is it for me to recover the data from your cloud service if I loss/forgot my password?"
On iDevice, it is NOT that hard for the phone owner to do it. It is trivial for any government agencies fake the owner's phone #, txt Msg, email to recover or reset the iCould password.
How do we know, and be sure that this is true?
The DUAL_EC_DRBG is a great example of how "golden keys" can be implemented securely, because the underlying problem is the same one that a lot of elliptic curve cryptosystems rely on.
Only the NSA could predict the output of the generator with the same level of security as ECDH or ECDSA.