This post is also available in:
עברית (Hebrew)
New research from Uruguay’s Universidad de la República Montevideo reveals that HDMI cables can be used to retrieve data shown on the computer screens they are connected to, and hackers might already know that.
In the paper, titled Deep-TEMPEST: Using Deep Learning to Eavesdrop on HDMI from its Unintended Electromagnetic Emanations, published on arXiv, the researchers explained that they were able to harness the power of AI to monitor LED screens connected to HDMI cables, previously deemed safe. The way they made it work was by, first, capturing the electromagnetic radiation the computer’s HDMI cable emits. Then, they trained an AI model by feeding it the data produced by a computer screen, along with the corresponding radiation signals that moved through the computer’s HDMI cable at the same time. After a while, the AI model was able to translate certain signals from the cable to the output showcased on the screen.
According to TechXplore, the team’s test yielded a reconstruction accuracy of 70% from a random computer screen. This is likely enough for understanding the main points of text displayed on the screen, as well as using this technology for more nefarious reasons, such as retrieving sensitive information, accessing passwords, credit card information, and possibly encrypted communications. The results can even be improved by running the deciphered output through a text recognition software, the researchers found.
As long as the HDMI cable is within a radius that the electromagnetic radiation emitted from it can be detected, and as long as that there is access to the technology used for interception, this technology can be used to decipher the data displayed on the screen. The researchers assess that hackers might already be aware of this, and that users might have already been targeted, unbeknownst to them. However, they claim that due to the unique expertise this technique requires, the likely targets are corporations and government entities, rather than the average individual.
