Thanks to the heat left by the fingers on the keys of a keyboard, an Artificial Intelligence is able to discover even the most complex passwords. Explanations.
Here is a new technique that should please the spy services. Provided you can gain access to the targeted device, it is possible to guess the passwords entered on a computer, a smartphone and especially an ATM. How ? By analyzing the heat residue left by the user’s fingers when he typed his password or code. It works with a keyboard or, of course, a screen and therefore also on the keys of an ATM.
This funny idea comes from researchers at the University of Glasgow, Scotland. They developed a system called ThermoSecure. Through this experiment, they sought to show that with thermal cameras whose prices are falling (approx. 250 euros) and artificial intelligence, sometimes freely available, a clever hacker can create a system at low cost to collect the identifiers of a device in seconds. They called this threat, “thermal attacks”.
For this to work, the user must have entered his password or his code a little before. It’s a bit strange to unlock a device and then leave it, but why not. On the other hand, this is necessarily the case for an ATM. The hacker must then use a thermal camera to take a picture of the keyboard or the screen. On the image, the clearer an area appears on the image thermal, the less time it has been hit. It is by examining these zones that it is possible to determine the keys, letters or symbols used, but also their order of entry. Even on a complex password, it only takes a few combination attempts to manage to decipher the sesame.
100% success for sesames of six characters
According to the researchers, even a neophyte who had been told how to decipher the thermal image could manage to find the password. However, the image must have been taken between 30 seconds and one minute after the surface was touched. But the researchers decided to go further and automate password discovery using a machine learning algorithm.
To feed it, they took 1,500 thermal photos of keyboards from different angles. Keyboards had just been used to type in passwords. It is with the help of probabilities that they were able to refine their model and achieve an efficiency of 86% in finding passwords, 20 seconds after entering them. The figure drops to 62% after one minute. We can say that the process is not necessarily efficient, but it should be noted that these figures correspond to long passwords of sixteen characters. When they do not exceed eight characters, the rate increases to 93%. With six, the algorithm consistently hits the mark. There is still a downside, because the effectiveness of the system will also depend on the material used for the keys of a keyboard.
This method, which nevertheless requires being in the immediate vicinity of the target, should not be applicable for long with the new protection systems without passwords which comply with the Fido protocol and which have just been implemented by Apple on its recent devices. In the future, other devices should also see this process. There remains the case of vending machines which could be the target of future hacker-thieves.