Google DeepMind Research Reveals AI Can Achieve 1,600 Times Speedup in Breaking Legacy RSA Encryption

·by Henderson·Engineering
Google DeepMind Research Reveals AI Can Achieve 1,600 Times Speedup in Breaking Legacy RSA Encryption
Key Points
  • Google DeepMind's research shows AI can accelerate the breaking of RSA encryption.
  • The study focused on 1024-bit RSA, an encryption standard that has been replaced by stronger 2048-bit keys.
  • AI can significantly speed up the evaluation of mathematical hypotheses, enhancing cryptanalysis efficiency.
  • The findings underscore the ongoing need to reassess encryption assumptions.

Researchers at Google DeepMind have discovered an encryption technique using artificial intelligence agents that reveals vulnerabilities in legacy RSA encryption at a speed 1,600 times faster than previous methods. This breakthrough was announced today by Pushmeet Kohli, Vice President and Chief Scientist at DeepMind, on social media. The research marks a significant advancement in the field of cryptanalysis without compromising modern encryption standards.

The study focused on 1024-bit RSA, an encryption standard that has been replaced by stronger 2048-bit or larger keys. According to Google DeepMind, these findings do not threaten modern encryption used to protect online banking, messaging applications, software updates, or digital currencies. Instead, they demonstrate how artificial intelligence can significantly accelerate the evaluation of mathematical hypotheses that underpin digital security.

Applications of AI in Cryptanalysis

Cryptanalysis is the science of testing encryption systems by looking for weaknesses, with the goal of preventing malicious actors from exploiting these vulnerabilities. Progress in this field has often taken years or even decades, allowing security experts to gradually strengthen encryption standards as computational power evolves. The company stated that its AI agents, under human supervision, compressed this timeline by rapidly exploring mathematical problems that traditionally required extensive manual research.

This research was led by Dj Dvijotham and Abhradeep Thakurta, in collaboration with researchers from the University of California, San Diego, and France's Inria, resulting in two major findings related to the legacy RSA-1024 cryptosystem.

The first finding pertains to digital signatures, which allow computers to verify that software updates, websites, and electronic documents come from legitimate sources. Digital signature schemes typically add extra preset information known as "padding" before signing data. Padding has long been thought to prevent forgery, but researchers had not fully quantified its effectiveness when attackers could influence parts of the signed message.

During the research process, the AI agents discovered that once an attacker controls about half or more of the signature data blocks, padding provides almost no additional protection. In this scenario, forging a digital signature mathematically becomes as simple as attacking an unpadded system. To validate this finding, researchers tested this method in a sandbox environment using previously solved RSA-1024 benchmarks from earlier academic work. The AI-generated implementation reproduced the result, reducing the required CPU time by a factor of 3.5, with computational demands less than one-thousandth of those needed to factor the benchmark key itself.

Even when an attacker controls only a small amount of signature data, such as a 32-byte digital fingerprint, the AI agents identified a new algorithm that theoretically improves the best attack on a 1024-bit RSA key by approximately 1,600 times.

Discovery and Impact of New Algorithms

The second breakthrough targets the batch factorization of multiple RSA keys of the same bit length. Since many RSA deployments use standardized key sizes, attackers could theoretically share the cost of expensive computations across multiple keys. The algorithm discovered by the AI improves the incremental efficiency of this process, reducing the average computational cost for each additional 1024-bit RSA key by about 30 times compared to previous batch techniques.

Kohli stated that these results highlight the ongoing need to reassess encryption assumptions as artificial intelligence becomes capable of tackling advanced mathematical problems. While current encryption standards remain secure, researchers believe that AI-driven cryptanalysis could become an important tool for stress-testing future encryption systems before emerging threats put them at risk.

Impact of AI-Driven Cryptanalysis on Security

As artificial intelligence technology advances, the efficiency of cryptanalysis has significantly improved, posing challenges to existing encryption standards. Although current encryption techniques remain secure, the research findings indicate that AI can uncover vulnerabilities in encryption systems in a short time, suggesting that more frequent reviews and updates of encryption assumptions may be necessary to address potential security risks. This research not only propels the development of cryptography but also underscores the importance of technological advancements in digital security.

H
About the author
Henderson