Academic researchers have developed a deep learning-based acoustic side-channel attack that can be used to steal data from keyboard keystrokes that are recorded using a nearby microphone or through the device’s own microphone. The trained model picked up recorded keystrokes with an accuracy of 95% when recorded from a smartphone microphone placed seventeen (17) centimeters away, 93% accuracy when recorded from Zoom, and 91.7% accuracy from Skype. This research emphasizes how sound-based side channel attacks may be a lot more feasible and dangerous than previously anticipated. The increased abundance of microphone-bearing devices with high-quality audio capture combined with ever escalating advancements in machine learning makes this statement especially true. The process works quickly by reordering the keystrokes on the target’s keyboard by either capturing sounds through a nearby microphone, infecting a device with malware that has access to its microphone, or potentially over an online meeting like a Zoom or Skype call. The recorded keystrokes are then transformed into waveforms and spectrograms to produce visualized identifiable images created by differing keystrokes that are then compared against the “CoAtNet” image classifier that was originally trained with sequences of keystroke recordings produced by the researchers. Readers who are potentially at risk of being targeted by this type of attack can mitigate it by using a software-based keystroke audio filter and continuing to use strong passwords as well as two-factor authentication.