How AI reads text via the electromagnetic radiation of HDMI cables

The new approach solves a long-standing obstacle to capturing digital video signals, which were considered too complex to reliably reconstruct due to their 10-bit encoding and nonlinear properties. The researchers therefore posed the problem as an inverse task and developed a deep learning model based on the DRUNet architecture. It converts electromagnetic signals back into a visual image. Another important advantage of the system is that it does not require complex physical measurements for each configuration, as it uses mathematical analysis and creates its own learning patterns.

Compared to older methods, which had an error rate of up to 90 percent, the new system reduces this to less than 30 percent. Despite significant progress, the results are far from perfect, as the reconstructed text still contains a significant amount of errors. This currently limits the possibility of direct malicious use in practice. The technology works with commonly available software-defined radio hardware and an extension for GNU Radio. The researchers have also made the source code and a database of approximately 7 gigabytes publicly available.

The discovery undoubtedly reopens the issue of screen security and electromagnetic data leakage. To reduce the risk, experts recommend using appropriate cables with better electromagnetic shielding, ferrite rings and specially certified housings. Ordinary cables and screens generally do not have the same level of protection against this type of interception.