"Proposed Illegal Image Detectors on Devices Are 'Easily Fooled'"

The use of built-in scanners on devices such as phones, tablets, and laptops to detect illegal images has been proposed by companies and governments. However, researchers from Imperial College London found that the proposed algorithms to detect such images on devices can easily be fooled with imperceptible changes to the images. Their findings were published as part of the USENIX Security Conference. The researchers tested the robustness of five similar algorithms and discovered that altering an illegal image's unique signature on a device can allow it to evade the detection algorithms almost 100 percent of the time. Their testing proved that perceptual hashing-based client-side scanning (PH-CSS) algorithms are not a magic bullet solution for detecting illegal content on personal devices. This finding raises questions about the effectiveness and proportionality of current plans to combat illegal material through on-device scanning. According to senior author Dr. Yves-Alexandre de Montjoye, of Imperial's Department of Computing and Data Science Institute, applying a specifically designed filter imperceptible to the human eye allowed the team to mislead the algorithm into perceiving two nearly identical images as different. The researchers' algorithm was able to generate various filters, which can make it difficult to develop countermeasures. Apple recently postponed plans to introduce PH-CSS on all its personal devices due to privacy concerns. Certain governments have also been reported to be considering using PH-CSS as a law enforcement technique. This article continues to discuss the study that found it possible to easily fool algorithms proposed to detect illegal images on devices and the concept of PH-CSS algorithms. 

Imperial College London reports "Proposed Illegal Image Detectors on Devices Are 'Easily Fooled'"