Google DeepMind Uses AI to Lower the Cost of Cracking Legacy RSA Encryption
Researchers at Google DeepMind have used reinforcement learning to find shortcuts in lattice reduction, reducing the computational cost needed to crack 1024-bit RSA encryption. Security experts are urging organizations to migrate to 2048-bit or higher standards as soon as possible. The breakthrough does not immediately threaten modern encryption but signals that older systems are increasingly at risk.
<p>Researchers at Google DeepMind have developed an AI-driven method that lowers the computational cost of cracking legacy 1024-bit RSA encryption, according to a report published this week.</p>
<p>The team used reinforcement learning agents to identify shortcuts in lattice reduction, a mathematical technique used in cryptanalysis. By finding more efficient paths through the computation, the researchers reduced the resources needed to factor large RSA keys, which is the core challenge in breaking RSA encryption.</p>
<p>Security experts say the breakthrough does not immediately threaten modern encryption systems, which typically use 2048-bit or 4096-bit RSA keys. However, they are urging organizations that still rely on 1024-bit keys, which were considered secure until the mid-2010s, to migrate to stronger standards immediately.</p>
<p>The findings add to a growing body of research suggesting that AI tools are accelerating progress in cryptanalysis. Earlier this year, separate research teams demonstrated AI-assisted attacks on other legacy encryption schemes.</p>
<p>The National Institute of Standards and Technology (NIST) has already published post-quantum cryptography standards designed to resist attacks from both classical and quantum computers. Security professionals say the DeepMind findings reinforce the urgency of adopting those standards.</p>
<p>Google DeepMind has shared the research with the broader security community and is working with standards bodies to assess the implications.</p>


