Earlier this week a young government contractor named Reality Winner was accused by police of leaking an internal NSA document to news outlet The Intercept. The documents outline the intelligence community's take on Russian efforts to hack a variety of companies responsible for facilitating US election voting. You can read the documents here.
Despite what anyone might have to say about the issue on Twitter, an arrest involving an accusation of any crime by any law enforcement agency in any country is not evidence of guilt. Even the most circumspect appraisal of the US justice system will reveal that tens of thousands of individuals are arrested every year only to have those charges *immediately* dismissed by a court, while nearly everyone who actually is *convicted* of a crime in this country has their charges reduced. Even in cases in which individuals have been convicted of the the most serious capitol crimes, courts have been forced to release dozens of individuals after DNA testing offered conclusive proof of innocence.
The point is this: being arrested is not being convicted. And being convicted is not proof-positive of guilt.
For the purposes of this post I will set aside the substance of the leak itself; again, I recommend reading the Intercept's initial reporting. This post is focused on reports of how law enforcement is claiming that it identified young Ms Winner and the consequences of these reports for computer users with an interest in privacy. The Electronic Frontier Foundation (EFF) describes the purported technique involved as follows:
The term for this technique is "forensic watermarking", "printer stenography" or "counterfeit deterrence system". The EFF definitively uncovered that a wide array of some of the most popular modern printers now print some form of watermark that can be used to definitively identify the device that printed a given document after a series of FOIA requests to some 10 US government agencies in 2008. The documents recovered through that FOIA request (some of whom date back to the 1990's) reveal that the watermarking techniques have been available since at least the 1980's, that printer manufacturers "voluntarily" adopted forensic watermarking under the ostensible justification of fighting counterfeiters, and that efforts to proliferate the use of watermarking involved the EU as well as the US.
The watermark involved in the documents published by the Intercept consists of a pattern of yellow dots that, when translated, identifies the serial number of the printer used and the date & time the document was printed. Here are those dots, made more visible by introducing additional contrast (images c/o Errata Sec's excellent post on this topic):
And here is the data gleaned from translating that watermark:
When this information is combined with the a standard corporate asset tagging system and printer logs, this watermarking can easily identify the workstation that printed a given document. The same technique can be used to create evidence that +a printer seized from a defendant's property generated a given document, as well.
So how does a privacy-conscience printer-user avoid this watermarking technique?
For one thing, do not assume that because you are unable to see any visible watermarking on documents from your printer that you are safe. Here is a photograph of a watermarked document taken in tandem with a Digital Blue QX5 microscope:
Even with the microscope the forensic dots are barely visible. Attempting to view the pattern without any form of artificial enhancement is a fool's errand.
A user can avoid purchasing one of the printers that EFF has tested and confirmed generates watermarks. Unfortunately, this list is not up-to-date; and as time goes on, the likelihood that *all* manufacturers will produce some form of watermarking increases.
The specific technique that involved the leaked documents published by the Intercept requires the use of color: the dots are a pale shade of yellow that is not easily visible without some form of digital enhancement. Avoiding the use of a color printer can avoid this specific technique. I am very skeptical of claims online that printing documents in "black & white" mode on color printers provides any form of protection: watermarks can easily be imprinted in greyscale (see "binary image watermarks"), and I have yet to see confirmation that this technique is anything other than *not* effective.
Even the use of a modern black and white printer leaves me uncomfortable. There are numerous means of imprinting imperceptible watermarks; the popular yellow dots are simply one technique of many. DCT based watermarking techniques are significantly more complex to identify than just adding some document contrast; until now the computational expense required for DCT was likely cost prohibitive for manufacturers. This is certain to change over time.
Tools designed specifically to protect users from this manner of technology are few to non-existent. I can't point the finger; I have not worked on this problem. I do have some ideas. Given that all water-marking techniques in use are unknown, it would likely be more reliable & perhaps cross-device-compatible to spoof identifying device information prior to reaching the printed document than attempting to identify & remove or modify the watermark itself. I have only marginal experience with peripheral firmware or drivers, but if anyone is interested in this type of project I learn fast & would be happy to help.
Despite what anyone might have to say about the issue on Twitter, an arrest involving an accusation of any crime by any law enforcement agency in any country is not evidence of guilt. Even the most circumspect appraisal of the US justice system will reveal that tens of thousands of individuals are arrested every year only to have those charges *immediately* dismissed by a court, while nearly everyone who actually is *convicted* of a crime in this country has their charges reduced. Even in cases in which individuals have been convicted of the the most serious capitol crimes, courts have been forced to release dozens of individuals after DNA testing offered conclusive proof of innocence.
The point is this: being arrested is not being convicted. And being convicted is not proof-positive of guilt.
For the purposes of this post I will set aside the substance of the leak itself; again, I recommend reading the Intercept's initial reporting. This post is focused on reports of how law enforcement is claiming that it identified young Ms Winner and the consequences of these reports for computer users with an interest in privacy. The Electronic Frontier Foundation (EFF) describes the purported technique involved as follows:
Imagine that every time you printed a document it automatically included a secret code that could be used to identify the printer - and potentially the person who used it. Sounds like something from an episode of "Alias" right? Unfortunately the scenario isn't fictional. In a purported effort to identify counterfeiters the US government has succeeded in persuading some color laser printer manufacturers to encode each page with identifying information. That means that without your knowledge or consent an act you assume is private could become public. A communication tool you're using in everyday life could become a tool for government surveillance. And what's worse there are no laws to prevent abuse.
The term for this technique is "forensic watermarking", "printer stenography" or "counterfeit deterrence system". The EFF definitively uncovered that a wide array of some of the most popular modern printers now print some form of watermark that can be used to definitively identify the device that printed a given document after a series of FOIA requests to some 10 US government agencies in 2008. The documents recovered through that FOIA request (some of whom date back to the 1990's) reveal that the watermarking techniques have been available since at least the 1980's, that printer manufacturers "voluntarily" adopted forensic watermarking under the ostensible justification of fighting counterfeiters, and that efforts to proliferate the use of watermarking involved the EU as well as the US.
The watermark involved in the documents published by the Intercept consists of a pattern of yellow dots that, when translated, identifies the serial number of the printer used and the date & time the document was printed. Here are those dots, made more visible by introducing additional contrast (images c/o Errata Sec's excellent post on this topic):
So how does a privacy-conscience printer-user avoid this watermarking technique?
For one thing, do not assume that because you are unable to see any visible watermarking on documents from your printer that you are safe. Here is a photograph of a watermarked document taken in tandem with a Digital Blue QX5 microscope:
Even with the microscope the forensic dots are barely visible. Attempting to view the pattern without any form of artificial enhancement is a fool's errand.
A user can avoid purchasing one of the printers that EFF has tested and confirmed generates watermarks. Unfortunately, this list is not up-to-date; and as time goes on, the likelihood that *all* manufacturers will produce some form of watermarking increases.
The specific technique that involved the leaked documents published by the Intercept requires the use of color: the dots are a pale shade of yellow that is not easily visible without some form of digital enhancement. Avoiding the use of a color printer can avoid this specific technique. I am very skeptical of claims online that printing documents in "black & white" mode on color printers provides any form of protection: watermarks can easily be imprinted in greyscale (see "binary image watermarks"), and I have yet to see confirmation that this technique is anything other than *not* effective.
Even the use of a modern black and white printer leaves me uncomfortable. There are numerous means of imprinting imperceptible watermarks; the popular yellow dots are simply one technique of many. DCT based watermarking techniques are significantly more complex to identify than just adding some document contrast; until now the computational expense required for DCT was likely cost prohibitive for manufacturers. This is certain to change over time.
Tools designed specifically to protect users from this manner of technology are few to non-existent. I can't point the finger; I have not worked on this problem. I do have some ideas. Given that all water-marking techniques in use are unknown, it would likely be more reliable & perhaps cross-device-compatible to spoof identifying device information prior to reaching the printed document than attempting to identify & remove or modify the watermark itself. I have only marginal experience with peripheral firmware or drivers, but if anyone is interested in this type of project I learn fast & would be happy to help.