DeepMind found RSA-1024 vulnerability 1600 times faster

Google DeepMind's AI agents found weaknesses in legacy RSA-1024 encryption 1600 times faster than previous methods. Modern keys are safe.

Researchers at Google DeepMind have carried out a critical study focusing on the RSA encryption infrastructure, one of the basic security protocols of the digital world. The developed autonomous artificial intelligence agents managed to detect structural vulnerabilities in the old generation 1024-bit RSA architecture 1600 times faster than the current state-of-the-art method. Although the findings have a wide impact in cryptography circles, modern encryption systems preferred in current banking infrastructures and instant communication platforms do not pose a direct risk.

Mathematical Weakness in the Digital Signature Mechanism

The technical basis of the research is the fixed additional data mechanism that is placed in the data series to be signed before the process in order to prevent fraud attempts. Mathematical analyzes carried out by artificial intelligence agents have proven that this defense layer completely loses its protection function if the attacker takes control of approximately half or more of the data block to be signed. When the specified conditions are met, creating a fake signature becomes mathematically no different from attacking a simple system that does not contain additional data. While the developed algorithm improved the known most effective attack model against 1024-bit keys by 1600 times, it reduced the processor usage time by 3.5 times in the isolated test environment. Moreover, this process consumed less than one-thousandth of the massive processing power needed to prime factorize the key itself.

Sectoral Reflections and Current Systems Status

Despite the high performance increase achieved, there is no immediate situation that will cause concern for the cyber security industry and end users. Because financial institutions, e-commerce platforms and messaging services have been using modern encryption standards at 2048 or 4096 bits instead of 1024-bit keys for a long time. Since the algorithmic approach developed by DeepMind cannot directly crack these new generation high-bit keys, the daily data security of individual users is protected. Nevertheless, this study shows how artificial intelligence-driven systems will be decisive in the speed of cryptanalysis and detection of security vulnerabilities in the near future.