DeepMind AI Exposes Legacy RSA Weaknesses 1,600 Times Faster
Automated cryptanalysis reveals structural vulnerabilities in 1024-bit digital signatures while leaving modern encryption intact.
On November 6, 2026, researchers at Google DeepMind revealed algorithmic techniques that accelerate mathematical attacks on older RSA encryption by up to 1,600 times. The study uses artificial intelligence to search for mathematical weaknesses in legacy cryptographic formulas, leaving modern 2048-bit encryption systems unaffected.
Bypassing digital signature padding
The research was led by Dj Dvijotham and Abhradeep Thakurta at Google DeepMind, working with cryptographers from the University of California San Diego and France's Inria. DeepMind Chief Scientist Pushmeet Kohli announced the findings on November 6, 2026, noting that the study evaluated the mathematical assumptions that protect digital records, as reported by Interesting Engineering.
Digital signatures rely on padding, which adds preset data to verify authentic software updates, secure websites and electronic documents. The team examined legacy 1024-bit RSA schemes to see how padding behaves during tampering. Autonomous software agents determined that when an attacker controls approximately half or more of a signed data block, padding provides virtually no additional protection against forgery. Under those conditions, forging a signature requires no more effort than attacking an unpadded system.
Testing benchmarks and batch factorization
Even when an attacker controls only a small fraction of the signed information, such as a 32-byte digital fingerprint, the newly identified algorithm theoretical efficiency improves on earlier attacks by roughly 1,600 times. To test the mathematical invention, researchers ran the routine in a sandbox environment against an established RSA-1024 benchmark.
The system reproduced the target result using 3.5 times less CPU time than previous methods. It also consumed less than one-thousandth of the computational power previously required to factor the benchmark key. A second technical finding focused on batch factorization, where attackers attempt to factor multiple keys of identical length simultaneously. The team discovered that their algorithm reduces the average computational cost of factoring each additional 1024-bit key by about 30 times compared with earlier batch techniques.
Future security stress tests
Modern digital security relies heavily on 2048-bit or larger RSA keys, which remain resistant to these findings. Current messaging applications, online banking platforms and operating system updates are not exposed to the 1024-bit vulnerability. The innovation instead demonstrates how automated tools can compress mathematical discovery cycles that historically took decades of manual research.
According to the DeepMind research team, supervised agents could become standard instruments for testing emerging cryptographic protocols before malicious actors find theoretical cracks. Security teams plan to use these automated cryptanalysis techniques to stress-test future post-quantum and public-key standards as computing hardware advances.
Read also on The Patent: AT&T and AST SpaceMobile complete first cellular to satellite network handoff · Humanoid Robot Learns Complex Sequences Without Unified Training Data
