AI Finds Encryption Flaw 1600 Times Faster Than Before

Google DeepMind team announced a new discovery in cryptanalysis and the team was led by Dj Dvijotham and Abhradeep Thakurta with researchers from University of California San Diego and Inria in France and the discovery was in old RSA encryption that uses a 1024 bit key and this encryption is old and no longer used in banks or modern apps that use 2048 bit or more but it still exists in old systems and AI found a new way to expose flaws in digital signatures 1600 times faster than the best method known before and digital signatures let computers check that software updates and websites and documents come from a clean source and AI found that if an attacker controls half of the signed data block or more then padding gives almost no extra protection and if an attacker controls only a small part like 32 bytes then AI found a new algorithm that improves the attack by about 1600 times for a 1024 bit key and it also found a way to improve analysis of a group of keys together and it cut the cost of analyzing each extra key by about 30 times and the link here is that the old idea relied on trying all chances by force while the new idea relies on AI learning the pattern of weakness and aiming the search directly at the weak point and this made cryptanalysis much faster and cheaper than before.

The similarity between the old way and the new way is that the goal is one and it is testing encryption power before anyone uses it badly while the difference is that the old tried everything and the new goes directly to the weak point and the scientific conclusion made is that AI does not break encryption by force but it shows the logic of weakness itself and that real safety is not in key length only but in padding design and data distribution and that any old system with a logical weak point will be exposed fast if a learning intelligence is aimed at .