Ladar claimed that he had no way to access user emails. The hidden subtext of that, which was sort of hand-waved away, was that he had no such access with the code he had written, and that all that stood between your mail and his eyes was his own willingness to write that code. The entire security of Lavabit depends on Ladar and his principles.
These systems are entirely built on code, and code is malleable. If you have control over the code that gets executed at every point in the stack (which the operator of a web app certainly does), there's always room for the operator to change the code and therefore change the behavior to do anything, including log things that were previously considered secure.
> The entire security of Lavabit depends on Ladar and his principles.
That has always been true. If he was a less scrupulous individual, he could have run an off-the-shelf e-mail solution with 0 fancy security, but claimed that it was all secure on the backend in his marketing. You wouldn't know the difference, because you can't audit the source code.
Unless you have the access, time, and expertise to perform your own source audit, any claims of security are always implicitly built upon a foundation of trust that the operator is doing what it claims.
"These systems are entirely built on code, and code is malleable. If you have control over the code that gets executed at every point in the stack (which the operator of a web app certainly does), there's always room for the operator to change the code and therefore change the behavior to do anything, including log things that were previously considered secure."
Which is exactly why nobody should have believed Ladar's claims in the first place. He trumpeted the fact that he had not yet developed any wiretapping capability and tried to distract everyone from the hard reality of "encrypted webmail."
This is why real security and privacy require end-to-end encryption.
You might be missing the point a little bit. You're correct, but end-to-end encryption is irrelevant if the operator changes the (e.g., Javascript) code in a webmail app and maybe even just for a single user.
The trust model has to include the device/OS/browser/etc that you're accessing the device on as well as all code and keys (including WOT servers, GPG keys, and the webmail code itself, or, if locally installed, all binary packages, updates, etc).
Which I think is the point is really that in the face of a government who can compel, e.g.,
* an OS vendor to install a backdoor in their software,
* a microprocessor vendor to critically weaken the Random Number Generator,
* the author and BDFL of a key operating system kernel to use said RNG,
* a distribution to include compromised or even unsigned packages,
* a mobile carrier who will allow random hardware on their network that can sniff, MITM, and inject evil data,
* a mobile phone manufacturer who will install CarrierIQ,
* (heavens!) a user/customer who visits an app store and installs an application with either a trojan attached or one that is directly a trojan,..
In other words, no amount of end-to-end crypto will save us if we're attacked at each end and in the middle by people who will stop at literally NOTHING (jail) just to get at the data. This is why Schneier wrote "The Government has betrayed the Internet."
Encryption is no panacea when you're dealing with someone who controls both ends and the middle. :( To put this on Lavabit is poor form, IMO. This is one end who stood up and fought and paid a huge price.
"You might be missing the point a little bit. You're correct, but end-to-end encryption is irrelevant if the operator changes the (e.g., Javascript) code in a webmail app and maybe even just for a single user."
Which is why, as I have said elsewhere, webmail is the problem here; more generally, you cannot have security or privacy as a service.
As for the government pushing back doors into products, that is at least upping the ante. It is harder to sneak back doors into software that people are actually using, especially software that is widely used, than it is to sneak a back door onto a server than only a handful of people can inspect. It is also harder to sneak in a back door when there are many, separately maintained implementations of a common protocol; you are forced to try to sneak a back door into an abstract description, which a lot of experts will be reviewing (see e.g. DUAL_EC_DBRG; the backdoor was discovered within a year of the PRNG being publicized).
There are also limits to what sort of back doors can be put at a lower level. Suppose you sneak a back door into my CPU, but I am running software that was developed after you developed your back door. You can make a straightforward computability argument that in general, your back door will be rendered useless by unexpected changes to software, particularly sweeping changes to it.
In any case, my point is not that encryption is a panacea, but rather that webmail is, by its nature, insecure. There was no way that Lavabit's security could have ever been more than just a handwave.
Ladar claimed that he had no way to access user emails. The hidden subtext of that, which was sort of hand-waved away, was that he had no such access with the code he had written, and that all that stood between your mail and his eyes was his own willingness to write that code. The entire security of Lavabit depends on Ladar and his principles.