DeepMind found RSA-1024 vulnerability 1600 times faster

RSA encryption has been the basis of secure connections on the Internet for years. Google DeepMind's artificial intelligence agents uncovered vulnerabilities in the older 1024-bit version of this system 1600 times faster than the best known method. Still, there's no need to worry about modern keys used in banking and messaging for now.

What Does the Vulnerabilities Found for RSA-1024 Change?

First of all, the main result of the study is about the digital signature mechanism. This mechanism adds a fixed additional data to the data to be signed before signing to prevent forgery. But AI agents have proven that this protection is useless if the attacker controls about half or more of the signed data block.

In such a case, forging the signature is not mathematically more difficult than attacking a system without additional data. Therefore, the protection layer does not provide the expected benefit in this scenario. The developed algorithm improves the best known attack on 1024-bit keys by 1600 times. Additionally, processor time decreased by 3.5 times in the isolated test environment.

Moreover, the application used less than one-thousandth of the computing power required to factor the key itself. Therefore, the result stands out not only in terms of speed but also in terms of resource usage.

Mass Factorization Method

The second breakthrough is on mass factorization. This method means calculating multiple keys of the same length simultaneously. Because many systems use standard key sizes. Therefore, attackers can share the heavy computational cost among several targets.

The algorithm proposed by neural network agents increased the asymptotic efficiency of this process. Asymptotic efficiency is a measure of how processing costs change as data grows. Thus, the average cost for each subsequent 1024-bit key dropped by a factor of 30 compared to previous techniques.

Is Modern Encryption Secure?

Yes, because the vulnerabilities found do not apply to keys of 2048 bits and above. These keys are used in online banking, messaging apps and digital currencies. Therefore, these areas are outside the scope of the study.

The study was conducted jointly with the University of California, San Diego and the French research center Inria. Additionally, the results were validated by reproducing the previously solved RSA-1024 benchmark in the sandbox. This testing showed that the methods were stable and could exceed manually imposed limits. Pushmeet Kohli, vice president and senior scientist of Google DeepMind, announced the finding.

According to the researchers, cryptographic assumptions need to be re-evaluated regularly. On the other hand, automatic cryptanalysis methods will also be a mandatory tool in stress testing future encryption standards. Thus, risks can be seen long before real threats emerge.