12 ms·

I was thinking along those Reality Winner lines that if I were an organization today getting a leaked document like this, what I would do would be to print it a

I was thinking along those Reality Winner lines that if I were an organization today getting a leaked document like this, what I would do would be to print it and then scan it grayscale and post that. Thereby being sure to avoid issues of embedded metadata and probably making it at least harder to read things like sneaky printer codes. Not that I'd really expect the Supreme Court to use something like that, but just as a general principle. Obviously this wouldn't be foolproof (something as simple as having a subtly different version of that office routing "stamp" on page 1 for different recipients would suffice to narrow things down a lot) but good practice anyway. Other than it being color, that's what I would have assumed was done here, except the staple holes and relatively decent image quality do seem to argue against it.

scarmig 4y ago
I've always thought that the best approach would be to embed some kind of steganographic identifier in the rendered text. Small variations in the spacing or even kerning. The simplest way to get around that would be to retype the whole document, which is a bunch of friction. But to counter that, more invasively, the texts different people would get could be different themselves. Sentences subtly reordered, slight differences in word choice or spelling. Though that makes collaboration more difficult. Less technical savvy but resourced organizations wouldn't face difficulty using those approaches. It could just be a bit that an IT admin toggles on the GSuite or MS Office admin panel.
- Manuel_D 4y ago
  You could use OCR, and reproduce it as all ASCII. Apparently EVE Online guilds would start to put minor variations in the text of memos, or even alter critical information (like changing timing of convoys or attacks by a few minutes) to weasel out informants.
  - jdougan 4y ago
    My facvorite was when the forum background image had the logged-in user encoded into it
- sidewndr46 4y ago
  Printers have done this since the 1980s https://en.wikipedia.org/wiki/Machine_Identification_Code https://en.wikipedia.org/wiki/Machine_Identification_Code
  - scarmig 4y ago
    That's the watermark, which is well known. Low resolution scans of the original document can counteract them (...I think...), so I'm brainstorming (dystopian) alternatives.
    - sidewndr46 4y ago
      I misunderstood the intent it seems.
- tablespoon 4y ago
  > Sentences subtly reordered, slight differences in word voice or spelling. Though that makes collaboration more difficult. I think that would make collaboration impossible. If that were standard practice collaboration would eventually uncover that was occurring, and a leaker would compensate. IMHO that technique is only valid when the recipients can be assumed to not be in contact, or as an unusual thing to identify a leaker when one is already suspected.
  - scarmig 4y ago
    > If that were standard practice collaboration would eventually uncover that was occurring, and a leaker would compensate. I don't think that's really the issue. Let's suppose you have 64 employees with access to a doc. You can uniquely identify them with 6 bits in the doc. A doc of a couple pages would be more than enough to put in those bits, and many more. Even if you had two collaborators who shared docs for a word for word diff, they'd still miss half the identifying bits, which reduces the potential culprit pool to 8, and that's assuming zero redundancy. The issue I imagine is more usability. Someone identifies one of those bits as content that should be changed: what happens when they suggest the edit?
    - tablespoon 4y ago
      > Even if you had two collaborators who shared docs for a word for word diff, they'd still miss half the identifying bits, which reduces the potential culprit pool to 8, and that's assuming zero redundancy. That's not really what I'm talking about. It seems like you're focusing on identifying all the exact bits that are being used to identify someone, but that's not important. What is important is the recipients figuring out that someone is monkeying with the documents in a systematic way to trace them. Once a leaker knows that, instead of leaking the exact document for publication, they'd either leak reworded summary of key points themselves, or require the journalist to only publish something similar.
dorfsmay 4y ago
Why not OCR the document to limit traceability?
- zerocrates 4y ago
  Sure you could do that too. You mean, just release it as text, I assume. It would certainly cut out a lot of possible methods of fingerprinting, leaving actual changes to the text only pretty much, which are probably not as common outside some limited scenarios. I think the reason you don't do that is that things like the formatting and so on being accurate lend credibility to the document.
  - dorfsmay 4y ago
    At the end of the day, credibility is a trust issue, and in the Photoshop era I don't think having the right header and the right formatting add much to the credibility.
- ghaff 4y ago
  Fewer potential clues but also less credibility I expect. OCR also isn't going to be 100% though probably more than good enough in this case. If you really want to implement subtle tracking, you could make changes in the document itself that differ by recipient.
MrMetlHed 4y ago
You either print and rescan and crop the edges, or you type it all out yourself. It'd be safest to never put it online at all other than paraphrased summaries. The next safest would be to just type it all out yourself in your own words, but it's pretty darn difficult to do when you've got legalese. Politico must have had reason to believe they're in the clear there. Maybe they had two copies or they knew their source got a shared copy that couldn't be traced. Journalists lose sleep over this.

scarmig 4y ago

I've always thought that the best approach would be to embed some kind of steganographic identifier in the rendered text. Small variations in the spacing or even kerning. The simplest way to get around that would be to retype the whole document, which is a bunch of friction. But to counter that, more invasively, the texts different people would get could be different themselves. Sentences subtly reordered, slight differences in word choice or spelling. Though that makes collaboration more difficult. Less technical savvy but resourced organizations wouldn't face difficulty using those approaches. It could just be a bit that an IT admin toggles on the GSuite or MS Office admin panel.

Manuel_D 4y ago
You could use OCR, and reproduce it as all ASCII. Apparently EVE Online guilds would start to put minor variations in the text of memos, or even alter critical information (like changing timing of convoys or attacks by a few minutes) to weasel out informants.
- jdougan 4y ago
  My facvorite was when the forum background image had the logged-in user encoded into it
sidewndr46 4y ago
Printers have done this since the 1980s https://en.wikipedia.org/wiki/Machine_Identification_Code https://en.wikipedia.org/wiki/Machine_Identification_Code
- scarmig 4y ago
  That's the watermark, which is well known. Low resolution scans of the original document can counteract them (...I think...), so I'm brainstorming (dystopian) alternatives.
  - sidewndr46 4y ago
    I misunderstood the intent it seems.
tablespoon 4y ago
> Sentences subtly reordered, slight differences in word voice or spelling. Though that makes collaboration more difficult. I think that would make collaboration impossible. If that were standard practice collaboration would eventually uncover that was occurring, and a leaker would compensate. IMHO that technique is only valid when the recipients can be assumed to not be in contact, or as an unusual thing to identify a leaker when one is already suspected.
- scarmig 4y ago
  > If that were standard practice collaboration would eventually uncover that was occurring, and a leaker would compensate. I don't think that's really the issue. Let's suppose you have 64 employees with access to a doc. You can uniquely identify them with 6 bits in the doc. A doc of a couple pages would be more than enough to put in those bits, and many more. Even if you had two collaborators who shared docs for a word for word diff, they'd still miss half the identifying bits, which reduces the potential culprit pool to 8, and that's assuming zero redundancy. The issue I imagine is more usability. Someone identifies one of those bits as content that should be changed: what happens when they suggest the edit?
  - tablespoon 4y ago
    > Even if you had two collaborators who shared docs for a word for word diff, they'd still miss half the identifying bits, which reduces the potential culprit pool to 8, and that's assuming zero redundancy. That's not really what I'm talking about. It seems like you're focusing on identifying all the exact bits that are being used to identify someone, but that's not important. What is important is the recipients figuring out that someone is monkeying with the documents in a systematic way to trace them. Once a leaker knows that, instead of leaking the exact document for publication, they'd either leak reworded summary of key points themselves, or require the journalist to only publish something similar.

dorfsmay 4y ago

Why not OCR the document to limit traceability?

zerocrates 4y ago
Sure you could do that too. You mean, just release it as text, I assume. It would certainly cut out a lot of possible methods of fingerprinting, leaving actual changes to the text only pretty much, which are probably not as common outside some limited scenarios. I think the reason you don't do that is that things like the formatting and so on being accurate lend credibility to the document.
- dorfsmay 4y ago
  At the end of the day, credibility is a trust issue, and in the Photoshop era I don't think having the right header and the right formatting add much to the credibility.
ghaff 4y ago
Fewer potential clues but also less credibility I expect. OCR also isn't going to be 100% though probably more than good enough in this case. If you really want to implement subtle tracking, you could make changes in the document itself that differ by recipient.

MrMetlHed 4y ago

You either print and rescan and crop the edges, or you type it all out yourself. It'd be safest to never put it online at all other than paraphrased summaries. The next safest would be to just type it all out yourself in your own words, but it's pretty darn difficult to do when you've got legalese. Politico must have had reason to believe they're in the clear there. Maybe they had two copies or they knew their source got a shared copy that couldn't be traced. Journalists lose sleep over this.