[perlpimp]

Sounds like a trap to me, not having the source code. Maybe it works for now but investing trust as things are now I'd rather go with opensource client.

Trust must be earned, so far it they brag about way they made tech work with a patched version of android - they don't really put forth anything that will give them credibility as a very secure protocol.

Claims without proof are just that.

[tptacek]

The track record of open source cryptography is bad.

[dublinben]

The track record of closed source cryptography is worse.

[IanChiles]

cough cryptocat cough

[atmosx]

The cryptocat issue can be viewed sideways:

* Open source applications are bad, see what happened to cryptocat?

* Open source is awesome! Look what happened to cryptocat!

If cryptocat was closed-source... would ever be noticed? I wonder...

[tptacek]

People insist on looking at this through their default prism of "closed source bad, open source good". But people with crypto experience have other prisms; for instance, "competent, well-vetted crypto" versus "amateur enthusiast crypto". Sometimes open source is also competent and well-vetted, but vetting is expensive, and there is a lot of amateur crypto out there.

[nocoment]

> Sometimes open source is also competent and well-vetted, but vetting is expensive, and there is a lot of amateur crypto out there.

You seem to be implying that one must be a hobbyist in order to write incompetent crypto software with no or incompetent review and tend to need company resources to get quality code reviews.

Having crypto is often an important checkmark and tack on for shipping a product and usually no one in the product group is competent to analyze the security of the way they tacked on encryption. If a few in the larger company are competent, they will avoid reviewing these projects. Being the engineer everyone associates with delays and frustrations doesn't do much for you and there will never be any proof of the costs you may have prevented.

The few better than I know how to criticize implementations that I have seen haveusually had considerable cross company and university involvement. That usually means open source or a lot of NDA and complex license agreements for cross organization code sharing.

[wglb]

For example?

[eliteraspberrie]

There are claims that Crypto AG products used to leak keys in ciphertext, at the request of the NSA:

http://www.hermetic.ch/crypto/kalliste/speccoll.htm

[tptacek]

I stifled the urge to say the same thing, but then realized that I'd lose the evening to defining what "mainstream" meant, after people dredged up random examples of snake oil from Schneier blog posts; not to mention the inevitable rehashing of the "beware custom algorithms and 390244 bit keys" thread, which is going to have to happen now because bringing up crypto truisms from the late-90s makes people feel smart.

[tga_d]

You're being unfair and you know it. Lavabit, the RSA fiasco, Apple's imessage crypto, etc. are all perfectly mainstream examples of closed-source crypto done wrong. As you said yourself, the only thing that conclusively makes a difference is if the crypto is "well vetted," having the source available is simply a means of making this easier. Classifying the quality of crypto-implementations based on the source model alone ("The track record of open source cryptography is bad.")is just disingenuous.

[tptacek]

No, I'm not being unfair. I don't think "open source" versus "closed source" has much at all to do with how secure a cryptosystem is. I do think having Trevor Perrin and Moxie Marlinspike working on your crypto design has a lot to do with how secure a cryptosystem is.

[uniclaude]

Please explain for those of us who are not good enough in the field (I'm genuinely asking).

I was under the impression that software like GnuPG and OpenSSL could be considered safe, so seeing a security professional warning about a negative track record of open source cryptography is worrisome.

What exactly should we be careful of when it comes to open source cryptography?

[lvh]

Not all open source code is broken; just a lot of it is. I think tptacek is trying to say that open source vs closed source is a mediocre predictor of the quality of a cryptosystem :)

[illicium]

OpenSSL