A team of UCB researchers have coupled the sound of typing to various artificial intelligence learning techniques and recovered the text that was being typed. This recalls to mind Peter Wright's work. Poking around the net, I found that Shamir and Tromer started from here:
Preceding modern computers, one may recall MI5's "ENGULF" technique (recounted in Peter Wright's book Spycatcher), whereby a phone tap was used to eavesdrop on the operation of an Egyptian embassy's Hagelin cipher machine, thereby recovering its secret key.
I haven't _Spycatcher_ to hand, but from memory the bug was set up by fiddling the phone in the same room to act as a microphone, and the different sounds of the typewriter keys hitting being pressed on the cipher machine were what allowed the secret key to be recovered. Here's some more of Wright's basic techniques:
One of Peter Wright's successes was in listening to (i.e. bugging) the actions of a mechanical cipher machine, in order to break their encryption. This operation was code-named ENGULF, and enabled MI5 to read the cipher of the Egyptian embassy in London at the time of the Suez crisis. Another cipher-reading operation, code-named STOCKADE, read the French embassy cipher by using the electro-magnetic echoes of the input teleprinter which appeared on the output of the cipher machine. Unfortunately, Wright says this operation "was a graphic illustration of the limitations of intelligence" - Britain was blocked by the French from joining the Common Market and no amount of bugging could change that outcome.
Particularly interesting is MI5's invention code-named RAFTER, which is used to detect the frequency a radio receiver is tuned to, by tracing emissions from the receiver's local oscillator circuit. RAFTER was used against the Soviet embassy and consulate in London to detect whether they were listening in to A4-watcher radios. Wright also used this technique to try to track down Soviet "illegals" (covert agents) in London who received their instructions by radio from the USSR.
Unlike Wright's techniques from the 60s, the UCB team and their forerunners have the ability to couple up their information to vastly more powerful processing (Ed Felton comments, the paper and pointer from Adam). They manage to show how not only can the technique extract pretty accurate text, it can do so after listening to only 10-15 minutes of typing without prior clues.
That's a pretty impressive achievement! Does this mean that next time a virus invades your PC, you also need to worry about whether it captures your microphone and starts listening to your password typing? No, it's still not that likely, as if the audio card can be grabbed your windows PC is probably "owned" already and the keyboard will be read directly. Mind you, the secure Mac that you use to do your online banking next to it might be in trouble :-)
While we are on the subject, Adam also points at (Bruce who points at) the CIA's Tolkachev case, the story of an agent who passed details on Russian avionics until caught in 1985 (and executed a year later for high treason).
The tradecraft information in there is pretty interesting. Oddly, for all their technical capability the thing that worked best was old-fashioned systems. At least the way the story reads, microfilm cameras, personal crypto-communicators and efforts to forge library passes all failed to make the grade and simpler systems were used:
In November 1981, Tolkachev was passed a commercially purchased shortwave radio and two one-time pads, with accompanying instructions, as part of an "Interim-One-Way Link" (IOWL) base-to-agent alternate communication system. He was also passed a demodulator unit, which was to be connected to the short wave radio when a message was to be received.
Tolkachev was directed to tune into a certain short wave frequency at specific times and days with his demodulator unit connected to his radio to capture the message being sent. Each broadcast lasted 10 minutes, which included the transmission of any live message as well as dummy messages. The agent could later break out the message by scrolling it out on the screen of the demodulator unit. The first three digits of the message would indicate whether a live message was included for him, in which case he would scroll out the message, contained in five-digit groups, and decode the message using his one-time pad. Using this system, Tolkachev could receive over 400 five-digit groups in any one message.
Tolkachev tried to use this IOWL system, but he later informed his case officer that he was unable to securely monitor these broadcasts at the times indicated (evening hours) because he had no privacy in his apartment. He also said that he could not adhere to a different evening broadcast schedule by waiting until his wife and son went to bed, because he always went to bed before they did.
As a result, the broadcasts were changed to the morning hours of certain workdays, during which Tolkachev would come home from work using a suitable pretext. This system also ran afoul of bad luck and Soviet security. Tolkachev's institute initiated new security procedures that made it virtually impossible for him to leave the office during work hours without written permission. In December 1982, Tolkachev returned his IOWL equipment, broadcast schedule, instructions, and one-time pad to his case officer. The CIA was never able to use this system to set up an unscheduled meeting with him.
Sounds like a familiar story! The most important of Kherchkoffs' 6 laws is that last one, which says that a crypto-system must be usable. The article also describes another paired device that could exchange encrypted messages over distances of a few hundred metres, with similar results (albeit with some successful message deliveries).Posted by iang at September 11, 2005 10:41 AM | TrackBack