College of Engineering News Room
USF Researchers Identify New Class of Attacks that can Eavesdrop Keystrokes
Associate Professor Yao Liu, in the Computer Science and Engineering Dept., along with student collaborators, published a paper at the 2018 ACM SIGSAC Conference on Computer and Communications Security in Toronto titled "No Training Hurdles: Fast Training-Agnostic Attacks to Infer Your Typing." This groundbreaking research reveals a new type of attack that can quickly infer what a user types without relying on training, which is a required step for traditional attacks.
Sensitive information such as classified documents, trade secrets, or private emails are typeset and input into a computer for storage or transmission almost exclusively via a keyboard. Security researchers at the University of South Florida have discovered a training-agnostic attack to eavesdrop what a user types.
“This attack eliminates the requirement to infect the target computer with a keylogger or other malware to violate the user’s privacy, nor does it require the line-of-sight between an adversary and the keyboard.” says Dr. Liu.
The attack stems from the ability of wireless signal to penetrate through obstacles. An adversary can usually conduct this attack by deploying a malicious 802.11 access point that provides free WiFi service to attract nearby victim computers, or utilizing a customized transmitter and receiver pair to launch this attack. For example, an adversary may attach a tiny wireless transmitter device to the back of the desk for eavesdropping what a user types. One important feature of this attack is that the adversary can infer typed content in a fast way through eliminating the training step, which is an essential step for traditional keyboard eavesdropping attacks.
The research team evaluated the impact of this attack and found that for a sample input of 150 words, the adversary can recognize an average 94.3% of these words within 1-2 minutes. The adversary is also able to decrease the entropy of a 9-character password from 54.8 bits to as low as 5.4 bits, vastly reducing the maximum brute-force attempts required for breaking the key from 31.08 quadrillion to just 42.