Google DeepMind: AI Discovers Two Vulnerabilities in Older 1024-Bit RSA Systems

On November 7, Google DeepMind disclosed that its use of human-guided AI agents in research uncovered two major vulnerabilities in the legacy 1024-bit RSA cryptosystem, reducing the computational cost required to break it below prior expectations.

Pushmeet Kohli, Vice President of Science at Google DeepMind, stated in a research note that one finding relates to digital signature forgery, while a newly developed algorithm can lower the average cost per additional key when batch factoring same-length keys. The study only impacts the deprecated 1024-bit RSA system, and modern cryptographic protocols using 2048-bit or longer keys remain secure.