Threat Actor: Hackers | hackers
Victim: Governments and large corporations | governments and large corporations
Price: Not specified
Exfiltrated Data Type: Screen data, encrypted messages, banking credentials
Key Points :
- Researchers in Uruguay developed an AI model to eavesdrop on screens via electromagnetic radiation leaks.
- The method can intercept signals from HDMI cables using an antenna placed outside a building.
- The eavesdropping process has a character error rate of about 30%, making it feasible for reading text.
- Hackers can capture signals even from outside a building or by installing devices to transmit data externally.
- Primary targets for this attack method are governments and large corporations, not home users or small businesses.
- Mitigation requires blocking electromagnetic waves from entire buildings.

Researchers in Uruguay have recently discovered a method to use artificial intelligence (AI) to eavesdrop on victims’ screens through electromagnetic radiation leaks from computer cables and monitors. This attack can be executed through various means, including the use of an antenna placed outside a building to intercept radiation signals from HDMI cables.
According to Federico Larroca at the University of the Republic in Uruguay, he and his team have developed an AI model capable of reconstructing digital signals from leaked signals at a distance of several meters. The National Security Agency (NSA) and NATO refer to these types of attacks as TEMPEST attacks, which involve intercepting and interpreting electromagnetic radiation signals that hackers can capture.
Using a method similar to that developed by Larroca’s team, hackers can spy on screens when users enter encrypted messages, banking credentials, or other personal information.
The researchers noted that hackers could intercept signals even when standing outside the building with an antenna. Malicious actors could also install a small device to capture signals, transmit data externally, and physically recover it.
The scientists trained the AI model with a set of original signals and intercepted signals. They then used text recognition software on the reconstructed images and compared them to the original screen images. The experimental results showed that the eavesdropping process produced an error rate of about 30% of characters, which is low enough for humans to accurately read most of the text.
Previously, advanced attack methods resulted in an error rate of over 60%. However, the research team believes that home computer users or small to medium-sized businesses should not be overly concerned about this.
According to Larroca, governments or large corporations are the primary targets for hackers using the aforementioned remote attack method. To mitigate the risk, entire buildings would need to block electromagnetic waves.
Related Posts:
Original Source: https://securityonline.info/ai-driven-tempest-attacks-uruguay-team-reveals-method-to-intercept-screen-data/