Researchers have flagged a new class of cyberattacks that use inaudible audio signals to manipulate AI systems and connected devices. These attacks embed hidden commands within sound frequencies beyond human hearing, making them difficult to detect.
This method uses ultrasonic or near-ultrasonic waves to transmit orders. While these waves are silent to humans, microphones installed in phones, computers, and smart speakers can easily sense them. Thereafter, an AI system recognizes these waves as voice orders and executes commands without the user’s awareness.
Voice assistant-enabled devices are at higher risk of falling victim to this kind of hack. This may involve mobile phones, smart speakers, or other IoT devices that constantly listen through microphones to execute tasks. The method can be conducted remotely when suitable conditions are available.
In addition to personal devices, other AI technologies that rely on voice-based command delivery can fall prey to this type of attack. With the increasing prevalence of AI technologies across many sectors, there is a growing risk of manipulation through hidden commands.
One of the biggest concerns is detection. The problem here is that users do not know an attack is underway because the signals are inaudible. Security systems today are not designed to detect ultrasonic input signals.
There are countermeasures, including a filtering process and an authentication layer, for executing voice commands. There have also been discussions about how to control command execution based only on valid inputs. The problem of inaudible signals represents yet another example of the growing threat from cybercriminals in line with technological progress.
Also read: OpenAI Takes on Anthropic with AI-Driven Cybersecurity Platform ‘Daybreak’