"Don't use the tool" is a cryptographically sound piece of advice, given what you know about the specific implementation, but it's utterly useless for Cryptocat, from a product standpoint. What's a developer supposed to do when a crypto expert just says, "no"? That's terrible feedback, it needs to be better, and it can be better (sometimes it actually is better, I think you've mentioned in the past that you've lent a hand to Cryptocat).
As with every other "theory vs. practice" discussion, the theorists need to tone down the rhetoric, and the practitioners need to actually listen to what the theorists are saying.
The right thing to happen with Cryptocat is for it to be scrapped. Its security issues pervade beyond bad cryptography. Its users are literally better off with Google Hangouts or iMessage.
And, in fact, the same was true of Lavabit: there was no recommendation that would allow a mail service that worked (for users) the same way Google Mail to resist serious attacks. But that was the premise of the system! It needed to be scrapped.
That's unambiguously bad, because the software industry thrives on failing a lot until a good solution is found.
There needs to be a way to develop crypto software that falls in a land of, "Don't use this yet, but maybe someday after we're done testing it, you can start using it and be secure" rather than "nope, just scrap the whole thing".
"the software industry thrives on failing a lot until a good solution is found"
may be at the heart of the disagreement. It is true that some parts of the software industry are best served by this style of iterative failure based progress, but it is unfair to say that the entirety of software is best developed this way.
There exist whole classes of problems where the failure modes of their solutions are quite simply worse than any potential value those solutions can provide. Those class of problems should NOT be iteratively developed, rather they should be rigorously engineered and formally verified.
I suspect your disagreement stems from whether you think crypto systems exist inside or outside that problem set.
> Joe Developer sees this advice and decides not to write crypto software.
False. Moe Manager stomps his foot and makes Joe Developer wire together "something good enough", regardless of Joe's qualifications. Since neither Joe nor Moe are able to tell how good a crypto system is, they invariably end up with something in the awful-to-kludge spectrum. Both Moe and Joe need to be repeatedly reminded that they don't know enough about this non trivial problem to make good enough calls.
> That's unambiguously bad, because the software industry thrives on failing a lot until a good solution is found.
"Unambiguosly bad" is a thought stopper, a rhetoric trick. This just means that since you don't like any of the options made available to you by reality (like... paying big bucks to a real expert), self deception is preferable.
Now, the fact that our software industry thrives on failure... I am ashamed to say that this just means our industry is really good at externalizing the consequences of our own incompetence to the customers. This is all good and fun, until real lives are on the line. Crypto is one of those cases, so it is one of those lines you don't cross.
> There needs to be a way to develop crypto software that falls in a land of, "Don't use this yet, but maybe someday after we're done testing it, you can start using it and be secure" rather than "nope, just scrap the whole thing".
Are you familiar with the term "Security by Design"? It is one of those pesky realities I was talking about. It basically says that you cannot create a product giving no concern to security, sprinkle some crypto fairy dust on top of it, and magically turn it into a secure product. If you try to do that, you'll end up with a product with broken security.
Which is not all that bad in practice. There are lots of useful products that we all are better of having in insecure form than not having at all. But if the whole point of product X is to be "Y, but done securely" there is no redeeming feature that can save X from the axe.
I guess I could point out that there lies, in a continuum of software, a place for experimental products which offer nothing other than the exploration of an idea and the opportunity to refine said idea through testing and discussion. I could also point out that your strawman of "sprinkle some crypto fairy dust" is entirely irrelevant to this conversation, and that your lack of understanding of how threat actors are modeled is evident in your inability to comprehend a tool that might protect against certain threats but can acceptably not protect against all threats.
But I think those points would fall on the "ears" of someone who just wants to be combative. Maybe someone else will pick up where you left off and engage in a more honest and less antagonizing manner, but this deep in the comment tree, "here be dragons".
Other than that, if you prefer to have non combative discussions, please stop bossing professionals around and tell them what to do. Specially if they are not providing a service you paid money for.
But it's funny, because what you're saying here (don't "boss" people around) is similar to what I am saying, with regard to experimental crypto software. The crypto experts need to let the developers develop, and find a constructive way to contribute to that process, rather than chicken-little every time a bug appears.
Imagine software that comes out, stating: "This software is not meant to do anything other than hide your porn from your mother. Please do not use this software to do anything other than protect yourself against a threat who has limited understanding of computers and software."
As that software matures, the disclaimer could expand to, "This software is not meant to do anything other than protect information against low-level criminal activity. Please do not use this software to do anything other than to protect your information from larceny or identity theft. This software will not protect you against a sophisticated or well-resourced actor."
Eventually expanding to, "This software has been tested by a community of crypto experts. While there is currently no known reason why this software won't protect your information from all bad actors, there is no guarantee in place that your information is absolutely safe. Always exercise extreme caution when protecting your data from sophisticated actors."
The idea being that, once an experimental piece of software becomes "good enough", it can be used to protect against a more diverse threat model.
The idea that you have to ship an NSA-proof product on day 1 is completely unfeasible and will never happen, and the crypto scientists out there need to realize that.
What I am trying to point out is that you cannot fix a flawed Security Design by incremental improvements. I am not talking about individual coding defects or even desirable features that have been left temporarily unimplemented on purpose. I am talking about Stupid!Ideas that get implemented because there was nobody around that recognized them as such, and then the usual traits of human nature (denial, previous investment bias, rationalization, etc) prevents the incumbents from scrapping those ideas until insurmountable evidence is assembled.
The article in question talks about using One Time Pads (OTPs) to conceal secrets. This is an idea that sounds good, OTP is mathematically proven to be "unbreakable" and all that... The problem is when theory meets practice and nobody asks the hard question: How are sender and receiver going to share the OTP prior to sending the real message.
This is not a trivial question to ask, giving the fact that OTP must have the same amount of bits as the message it is protecting. You can also not "recycle" a single OTP for many messages (that would be a Many Time Pad, which by definition violates the very preconditions that make OTP secure).
So, you end up with the scheme used in spy movies (I have no idea how real spies worked during the Cold War). Each spy carries around a code-book of OTPs that he received in hand by a trusted contact. The fact of being in possession of such book is itself a proof of being a spy so he was to take great care to hide it, which is hard because it is a whole book (OTP, lots a lots of messages). Also, the spy agency will have the cumbersome work of providing a different code-book to each spy (because otherwise, when the enemy catches one spy, they could theoretically run the OTP on all previous communications from suspected spies and have each one of them rounded and shot before the word comes out that the first spy was in prison).
This example also serves to illustrate another point. Maybe a Bad!Idea can be made to work in a limited way, if need is dire and resources are plenty. But any organization which kept secrets of real importance and tried to rely on this kind of solution would end up eaten alive by any adversary that had heard the words "asymmetric cryptography" spoken together.
I really can't find in anything I've written where I talked about writing a "bad" piece of crypto software. In fact, I've been talking about a potentially "good" piece of crypto software that isn't getting written because no software is perfect, but that's what crypto experts are demanding out of the gate.
Can you more succinctly sum up what you're trying to say? Do you think software shouldn't be written in the crypto space unless it's perfect from v0.0.0? I can't imagine you think that, so I'm wondering what exactly it is you're arguing here.
Except where it is important. Failing a lot in crypto can have deadly consequences that are not visible until it is too late.
He probably shouldn't. That's pretty much the point.
(--edit-- or he should consult with people who have studied this, find out about best practice, use existing solutions where possible, etc etc)
This is the problem -- you and others are trying to perpetuate the idea that software must always be "complete" the moment it's put on GitHub. Why must Joe Developer adhere to your rigorous definition of what a software product is?
Joe Developer should be free to tinker, and if he wants to tinker with crypto, crypto experts should let him. Currently they don't, and that's why we don't have any good crypto software.
He can tinker.
He can learn.
He can do any of the many excellent free courses, tutorials and challenges on the net.
He can implement attacks against toy problems, and real problems.
And eventually, he can get deeper into implementing it.
What he will absolutely and rightly be shot down for is making unverifiable claims about the security or what he's tinkering with, and try to profit from his claims.
I'm sorry if you have a problem with that.
It's a nice straw man, but not what I suggested at all.
When he launched an app that claimed to be the new safest thing ever, surely? Perhaps with an unbreakable One-Time-Pad. You know, the stuff we're commenting on here...
Nope, not at all the stuff we're commenting on. We're commenting on the stuff that hasn't been written yet and probably won't be written at all because of your negative attitude.
I'm not sure what it is you want here. Nobody's telling you not to use it. Nobody's telling you not to tinker. Everyone's encouraging you to learn, and giving you advice on the best way of doing that. You seem to want every bad, amateurish attempt at novel crypto, accompanied by self aggrandising copy, to be hailed as if it was awesome. It's not.
Again, you strawman me. Where in my argument did I talk about making outrageous claims? I would very much appreciate it if you separated "make outrageous claims" from "develop open source security software".
You appear to be annoyed that folks write take-downs on software like the one linked to by the article. This is the outrageous claim. If you feel this is not what you're saying then fine, I'm not sure we actually have an argument.
>> I would very much appreciate it if you separated "make outrageous claims" from "develop open source security software".
There's no reason not to develop OSS security software. There's every reason to put up big warnings saying you can't vouch for its security. There's every reason to ask people who know what they're talking about to take a look, and to be humble when they point out flaws, and to rework it if that's the case. There's every reason to take some of the excellent free courses.
But there's also no reason to go off-piste. Why develop this crazy new OTP scheme? There's existing crypto for exactly these use cases. Leave the devising and proving of new crypto constructs to the deeply-versed and the academics, who know how to prove this stuff (and disprove it).
It's like saying "I can't write my OSS networking project because people keep telling me not to rewrite the TCP stack without learning about TCP first". Do you need to rewrite the stack? If so, if you want to do a good job then it's going to take a hell of a lot of knowledge and training. Why not use the OS API?
I don't know what you think I wrote, but no one here at any time suggested writing a new crypto scheme.
It's 4 days later. I've made my point over and over again, and you understand it. You have even restated my point yourself, in your second paragraph.
I'm done. You agreed with me, and everything else you're trying to talk about is just you begging to be validated. I can't feed this anymore, sorry.
Can I ask you, when I say - "people writing software products generally shouldn't be creating novel cryptosystems" - what is it you hear?
Do you hear someone saying "you shouldn't be writing software that uses encryption" ?
Because that's not what I'm saying.
A cryptosystem is something like what's going on here - we have messages secured by OTPs, generated using SecureRandom and exchanged over an encrypted channel using an AES key displayed on the screen for the other party to scan.
So what I'm saying is "hey, why not use the industry standards for your crypto? Inventing cryptosystems is way hard and you probably got it wrong. You can still make your cool apps on top of it!"
Security people freak out at new amateur encryption tools because when they don't, those tools get adopted by hapless users who entrust real secrets with them. It's not a theoretical concern; it happens over and over again in the real world. It happened to Snowden twice, once with Cryptocat and once with Lavabit. In both instances, crypto experts begged people not to use those tools. In both instances, misguided privacy advocates pushed back on the crypto experts. The tools either didn't get fixed (because, in Lavabit's case, they couldn't be) or didn't get fixed in time. Almost certainly, NSA has decrypted intercepts as a result.
It's happening right now with other tools. It will, with certainty:100%, happen again with Cryptocat.
During the snake oil era of medicine, people used to take radium tonics as a cure-all. That's what these tools are: radium tonics.