OPSEC is hard.
I'm not sure how we could explain to avoid it - where would the explanation go? Visiting that page would be just as much of a correlation, no? It's kind of a chicken and egg problem, unless the source is already using Tor.
Avoiding the "trail of the SSL connection" also suggests we should be doing something to combat website fingerprinting, which we have discussed but do not have a clear solution for yet.
Our current thinking is that just visiting the landing page is not enough to prosecute a source. We can do better, and are working on it, but it's difficult.
You could put the instructions on pages that many people visit regularly, true security through obscurity. For example, put the instructions in abbreviated form in a box in the footer of your front page (or in the footer of every page).
Print it in the physical newspaper. The German computer magazine c't prints their PGP key fingerprints in the masthead.
1. Make the entire site available under `ssl.washingtonpost.com` (ideally without the `.ssl` prefix).
That way, the domain won't be as suspicious as it is right now. I suspect that this is more or less the only content hosted on the domain.
2. Include an iframe for all (or a random subset of) visitors, loading this particular url (hidden).
By artificially generating traffic to the endpoint it will be harder to distinguish these from other, 'real' requests.
Use a random delay for adding the iframe (otherwise the 'pairing' with the initial http request may distinguish this traffic).
3. Print the link, url and info block on the dead trees (the paper), as other has suggested.
4. Add HSTS headers (http://en.wikipedia.org/wiki/HTTP_Strict_Transport_Security)
We're working on a best practices guide for deployments [0]. I'll make sure these suggestions go in there. Feel free to take a look and comment if you're interested!
[0] https://securedrop.hackpad.com/SecureDrop-Deployment-Best-Pr...
Or, since the content of this page is mostly text, it could be included in the HTML of all washingtonpost.com home page requests with very small overhead, and shown with a non-tracked javascript action (link/button), so it is all client-side and indistinguishable from a normal request to the home page.
This may be one of the rare cases where the use of a QR Code is justified.
I agree it would probably be a good idea to put a warning about such a problem though.
Ultimately, we can write descriptive documentation - but getting it read and understood is hard. Cryptoparties, are again a great idea, but getting the non-technical user involved is damned hard.
IMHO these things always come down to "how do we make it easy for the public, whilst keeping it REALLY secure". How does security become a general piece of education, much akin to math, or at least history?
That should make it hard enough to correlate any data, I guess they have enough visitors.
You only need the two "leak at time X, IP Y loaded this page at time X-5" datapoints to break this.
An embedded page is not fetched by someone else.
My suggestion is to embed an iframe to the posted URL on every page on www.washingtonpost.com. Every article, everything. I'd assume this would blast the logs enough that if you look at "time X-5" you'll have too many data points to actually make something out of it. Because everyone who reads an article on wapo will have also visited that page. So yes, that embedded page would be loaded by every single viewer of any page on washingtonpost.com.
Edit: I just realized that there is a huge unfixable flaw in this approach. The request for an article in the logs will always show up shortly before the request for the SecureDrop page. Even if you would iframe a random article on the SecureDrop page too you could see from the logs that is was loaded before the actual article. Essentially rendering this thing useless :/
So... Nervermind... I guess.
I still believe it would not be enough, since such a thing could be silently disabled by WaPo if ordered to do so.
The point in SecureDrop is that they could not deanonymize the source even if they tried.
Generally, it is very difficult to convince the operators of sites like the Washington Post to do things like this, but we're working on it!
Other possible approach: load the landing page everywhere and show it with Javascript when the user clicks their way to it. I think it's an improvement on the iframe without drawbacks. How does it sound?
It's a hard problem :/
I guess you could randomize if you load the iframe or not. Then you couldn't be sure if a visit was an actual visit or an iframe that was randomly triggered (with a random delay).
But for this to be useful you'd still need to instruct sources to randomly browse the page before going to SecureDrop. Which might work if you force them to click a link on the main-page to get to the SecureDrop page.
But if they go directly to /securedrop it will fail again because the GET /securedrop will show up as the first request from that IP, giving away that the visit was intentional.
So my current idea would be to randomly generate the actual /securedrop path in a non-predictable matter per client. Maybe something simple like securedrop-sha1(...). Then link to that from WaPo's main page. Forcing everyone to go trough WaPo.com. But then you still have the problem that you must make sure sources don't access this link from history or something.
Quite a lot of work, for still flawed security.
I don't think you can connect visiting the info page and the very next SecureDrop file upload.
And "a group of 100 IPs including a coffee shop near NSA employee John Smith's home" is enough.
* Your actions could put you on a shortlist of people to be more thoroughly investigated.
* Your actions could tip off the people whom your information threatens; maybe they stop communicating with you (or worse) to shut off the leak.
* Per the Snowden release, the NSA tracked the communications of people within something like 3 degrees of their targets. With standards that low, it's not a stretch to think someone would track everyone visiting the Washington Post's secure drop box.
That wasn't proof, of course, but it didn't need to be proof, just a good lead for law enforcement to kick-start their investigation.
I wrote up an analysis of exactly this problem last year: http://grugq.github.io/blog/2013/12/21/in-search-of-opsec-ma...