NSA Redactions Fail, Fail and Fail Again.pdf

(1042 KB) Pobierz
In light of the recent kerfuffle with the New York Times' editorial blunder, which, among other things
exposed the name of an NSA employee. I'd like to spend a moment looking at other failures of
redaction. It is not true that redactions always fail, but they quite often do, for a variety of reasons.
They include technical glitches, accidental releases of unredacted version, sloppy blackouts, the
existence of different versions of the same document with different redactions, and the availability of
information that allows the backouts to be logicked out.
Technical Glitches:
The use of Adobe Acrobat is the biggest source of technical reasons that redaction fails. When a
document has been OCR'd and then blackouts added, the OCR version often remains. This is what
happened to the New York Times. A different PDF problem happened to the Washington Post, with
images, the editorial department had simply used Acrobat to draw white boxes over an underlying
image. The image could simply be copied from the source document, and pasted into a photo editor,
and the unredacted version would magically appear.
Accidental Release:
Der Spiegel briefly published an unredacted version, listing all the Special Collection Service (SCS)
sites. This was quickly removed from ready view on the website, but not before it had been copied on
to a Cypherpunks site, and later on Cryptome. Another version of this happens when an item meant to
be censored was not, such as a single instance of Google, in a court document.
Imperfect Blackouts:
Sometimes the blackouts are misaligned and the tops or bottoms of the words remain visible, this
happened to l'Espresso, again with another document listing SCS sites, thus allowing Amman, Jakarta
Riyadh, and Taipei to be read. A second version happens when text is simply blurred, this is not always
fully effective.
Existence of Sufficient Information to Figure Out the Redactions:
Otherwise known as “How dumb do you think we are?” Yet another document listing all the SCS sites
was published by NRC, with approximately half the names blacked out. But each of the SCS sites was
marked by a red dot on a map, allowing a sufficiently determined person (namely me) to work out the
remainder, by simply reading a map.
A second version of this happens when the blacked out portions are sufficiently well described in
articles to allow the blackouts to be revealed. In some cases, you can even reconstruct whole
documents solely from the descriptions.
Existence of Alternate Versions :
There are all kinds of variations on this theme: the same document can be FOIA'd at different times,
and come back with different redactions, the versions can then be compared to allow a much more
complete version to become public. In the case of leaks, one version has been shown on TV, and
another released online. There are several instances of this, O Globo published online a heavily
redacted document, that they had already shown in full on television. So had CBC.
1251386686.036.png
Technical Glitches
FASCIA slide as intended to be released by
Washington Post.
FASCIA slide revealed by copying the image
stored within the PDF and pasting it into a photo
editor. This version shows the limits of the
FASCIA database and NSA's MDR2 cloud on
which it resides.
Accidental Releases
Heavily redacted SCS site list as intended to be
released by Der Spiegel
Fully unredacted version as briefly shown on Der
Spiegel, before being removed from search,
though still available through direct link.
1251386686.037.png 1251386686.038.png 1251386686.039.png 1251386686.001.png 1251386686.002.png 1251386686.003.png 1251386686.004.png 1251386686.005.png 1251386686.006.png 1251386686.007.png 1251386686.008.png 1251386686.009.png 1251386686.010.png 1251386686.011.png
 
Imperfect Blackouts
SCS table as shown by l'Espresso. Image with the flaws in the blackouts highlighted.
Knowing that the list is alphabetized, and mostly
consists of world capitals, allows what few clues
there are to be exploited.
Existence of Sufficient Information to Figure Out the Redactions
SCS table, as shown by NRC The little red dots allow most of the sites to be
determined, or approximated. I have omitted some
deliberately, others are simply in regions that are
too crowded. The cities aren't necessarily in their
original slot in the legend.
1251386686.012.png 1251386686.013.png 1251386686.014.png 1251386686.015.png 1251386686.016.png 1251386686.017.png 1251386686.018.png 1251386686.019.png 1251386686.020.png 1251386686.021.png 1251386686.022.png 1251386686.023.png 1251386686.024.png 1251386686.025.png 1251386686.026.png
 
Existence of Alternate Versions
Slide as shown on O Globo's website
Slide as shown on Fantastico TV (composite of
close-up pans) Still a bit difficult to read.
1251386686.027.png 1251386686.028.png 1251386686.029.png 1251386686.030.png 1251386686.031.png 1251386686.032.png 1251386686.033.png 1251386686.034.png 1251386686.035.png
 
Zgłoś jeśli naruszono regulamin