"Decryption" of Bush Memo

jjq writes "A decryption of the so-called Bush Memo, see the CNN news, http://edition.cnn.com/2004/ALLPOLITICS/04/10/bush .briefing/ was presented at Eurocrypt 2004 (Interlaken) during the rump session. David Naccache shown how to recover words from removed text in several memos. See more about this story at Lemonde.fr (sorry, it is written in French)." Just use the the fish.

Comments

google translation - have fun.

[krymsin01]

Cryptologists decipher a term censured in a "memo" addressed by the CIA to Geoges Bush
THE WORLD|07.05.04| 12h58 UPDATED 07.05.04|16h13
Les annales et corrig÷Ydu baccalaurïE depuis 1995.
Abonnez-vous au Monde.fr, 5? par mois

A passage covered with black ink in a document recently diffused by the White House was reconstituted. The method could be applied to considerable d¦classifi¦es files.
It "was bored" in front of television, the weekend of Easter, "when the memo of the CIA with George Bush was diffused" , remembers David Naccache, specialist in the coding of the data of the French company Gemplus. "I telephoned at once Claire Whelan, a coed of Dublin City University, from which I direct the thesis, to propose to him to attack the caviard¦s passages" , it tells. Mission accomplished, or almost.

The "memo" in question, addressed on August 6, 2001 by the CIA to president Bush and entitled "Ben Laden determined to strike in the USA", had been just d¦classifi¦ by the White House. This one wanted to prove that the precision of the warnings of the services of information was not sufficient to make it possible to the president to prevent the attacks of September 11. But five passages specifying the sources of the collected information had been covered with black ink.

For the cryptologist David Naccache, these illegible fragments were as many red rags. The result of its efforts - "conduits on a purely deprived basis" , specifies it, concerned not to imply its employer in his initiative - was presented Tuesday May 4 at the time of the conference Eurocrypt 2004 which joined together until May 6 with Interlaken, in Switzerland, the gratin of world cryptography. "the demonstration was extremely impressive" , judge Jean-Jacques Quisquater (university of Leuwen-the-New), specialist in the field, which greets this company of " reverse engineering of censured document" .

David Naccache and his pupil indeed succeeded in discovering one of the censured words. The term "Egyptian" seems the only possible one to them. They want to polish their method before returning their verdict on a longer passage, in order not to discredit it. And they straightforwardly threw sponge for a completely isolated word, for lack of sufficient indices.

Technology employed does not have, at first sight, anything revolutionist. The two researchers initially "rectified" the text, deformed at the time of its digitalization - the slope was only of 0,52ã. They then used a software of character recognition to determine the police force of the text which fixes the number of signs per unit of length. The simple recourse to an English dictionary then makes it possible to draw up a list of possible words. "1 530 corresponded" , indicates David Naccache.

But the article "year" preceding the word mystery implied that this one necessarily started with a vowel, which made it possible to bring back the list to 346 words. In French, an index provided by articles like "one" or "one", in the same way, would have made it possible to tighten research. The selection was also facilitated by the fact that the bill of character, Arial, is "proportional", i.e. the "hunting" of the letters varies. The space occupied by an I differs from that taken by W, which can give additional indices, compared to the police forces known as "monospace", like the Mail, often used, where all the letters are worth.

"Among" the surviving "words, five or six could make direction, but only Egyptian corresponded to the context" , indicates the cryptologist. This last stage raises more human intelligence than of the geometry of the text. To choose among Ukrainian, univited, unofficial, incursive, Egyptian, indebted and Ugandan, the two researchers were based on their good direction, Uganda and the Ukraine seeming too far away from the theatre of the operations to be retained, for example.

No doubt the analysis of the "memo" of the CIA reveals only one "an open secre

Re:google translation - have fun.

[m000]

They showed that the second of the three blacked-out sections of the released portion of the PDB was most likely "Egyptian."

In this particular case, the section was relatively short, so it wasn't tough to come up with a possible word list, but I wonder how well they'll do on the other sections, which are longer and therefore more likely to be multiple words. In the end, they rely on context to decide which of the possible strings is most likely the correct, which even in this case allowed for some close possiblities. A member of Egyptian Islamic Jihad would probably be most likely to interact with a member of an Egyptian service, but could just as well have spoken to a member of an unofficial (e.g., news) service.

Re:google translation - have fun.

[Anonymous Coward]

I've been to Eurocrypt where this was presented, the answer is simple: they exhaust all words and then they wish to feed the exhausted sentences one by one into a syntaxical analyzer to rule out these sentences that make no sense in English. They recognize that the method will disclose one-word hidden words, will sometimes find two-word combinations but they think that three-word combinations is probably the limit.

Beyond attacking this memo the method simply points to the fact that the way in which we censor documents before release should be changed.

BTW this presentation won the "best rump session award"... a cow-bell.

Re:google translation - have fun.

[Peter Desnoyers]

Here's a cleaned up version of the translation. My French isn't great, but it's better than the fish. (although it's great having a computer do the grunt work...)

A passage covered with black ink in a document recently release by the White House was reconstituted. The method could be applied to a considerable number of declassified files.

He "was bored" in front of television, the weekend of Easter, "when the memo from the CIA to George Bush was released" , remembers David Naccache, a cryptographer at the French company Gemplus. "I immediately telephoned Claire Whelan, a student at Dublin City University, whose thesis I supervise, to propose to her to attack the blacked-out passages" , he relates. Mission accomplished, or almost.

The memo in question, sent on August 6, 2001 by the CIA to President Bush and titled "Bin Laden determined to strike in US", had just been declassified by the White House, which wanted to prove that the accuracy of warnings from intelligence services had not been sufficient to allow the president to prevent the attacks of September 11. But five passages specifying the sources of the collected information had been covered with black ink.

For the cryptologist David Naccache, these illegible fragments were red flags. The result of his efforts - "conducted on a purely private basis" , he specifies, concerned not to implicate his employer in his initiative - was presented Tuesday May 4 at the Eurocrypt 2004 conference, which met until May 6 in Interlaken, Switzerland, the gratin (???) of world cryptography. "The demonstration was extremely impressive" , judged Jean-Jacques Quisquater (University of Louvain-la-Neuve), specialist in the field, who applauds this technique of "reverse engineering censored documents".

David Naccache and his student indeed succeeded in discovering one of the censured words. The term "Egyptian" seems the only possible one to them. They want to fine-tune their method before rendering a verdict on a longer passage, in order not to discredit(???) it. And they totally threw in the sponge on a completely isolated word, for lack of sufficient indices.

The technology employed is not, at first glance, revolutionary. The two researchers initially straightened the text, deformed at the time of its digitalization - the slope was only 0.52 degrees. They then used character recognition software to determine the point size of the text which determines the number of characters per unit length. A simple recourse to an English dictionary then makes it possible to draw up a list of possible words. "1,530 matched" , indicates David Naccache.

But the article "an" preceding the mystery word, implied that this one necessarily started with a vowel, which made it possible to reduce the list to 346 words. In French, an index provided by articles like "un" or "une", in the same way, would have made it possible to tighten the search. The selection was also aided by the fact that the character font, Arial, is "proportional", i.e. the width of the letters varies. The space occupied by an I differs from that taken by W, which can give additional indices, compared to the fonts known as "monospace", like Courier, often used, where all the letters are the same width.

"Among the remaining words, five or six could fit, but only Egyptian corresponded to the context" , indicates the cryptologist. This last stage relies more on human intelligence than the geometry of the text. To choose among "Ukrainian", "uninvited", "unofficial", "incursive", "Egyptian", "indebted" and "Ugandan", the two researchers used their common sense. Uganda and the Ukraine seemed too far away from the theatre of the operations to be retained, for example.

No doubt the analysis of the CIA memo of the CIA reveals only an open secret, recognizes David Naccache. But the method automates the research. (???) In another memo, it revealed that civil helicopters militarized by the Iraqis had been bought in South Korea. And nothing prevents the automated application of this technique to an group of declassified documents, which might one day allow it to uncover "isolated words, or even groups of two or three words".

Direct Babelfish link

[Anonymous Coward]

Babelfish link

The technique

[Anonymous Coward]

The memo was written ina proportional font. It's a standard MS Windows font, so he can easily find the exact width of each letter (and kern pairs and the like). By looking at surrounding words, and allowing for the justification stretch on the line, you can figure out the exact width of the blacked-out words.

Then it's just a matter of doing a dictionary search to find which words could possibly fit, and use context information to figure out which one of the possibles it is.

Re:The technique

[belmolis]

To fill in a detail, they made use of the fact that the blacked out word is preceded by "an", from which they deduced that it had to begin with a vowel. The Babelfish mistranslation of this as "year" is therefore particularly misleading.

Re: Just remember...

[theglassishalf]

In LA, (which has a high proprtion of traffic deaths) there are 7.2 traffic deaths per 100,000 people per year.

130,000 Iraq troops, 721 dead in just over a year.

Yeah....