Google DeepMind announced on November 6, 2026 that its AI models, working under human direction, uncovered two distinct weaknesses in legacy 1024-bit RSA encryption.

According to Crypto Briefing, the first weakness enables the forgery of digital signatures, and the second is a new algorithm that lowers the cost of batch factoring for related moduli.

RSA security rests on the difficulty of working backward from a product of two large primes to find those primes, and batch factoring attacks many moduli at once when they are related in certain ways.

The work was led by DeepMind researchers Dj Dvijotham and Abhradeep Thakurta, collaborating with cryptography experts Nadia Heninger and Emmanuel Thomé.

The bigger surprise is tempo rather than the flaws themselves. Crypto Briefing reports that AI agents can significantly cut the time needed to probe the mathematical assumptions encryption depends on, work that historically could take years of dedicated human effort.

Pushmeet Kohli of Google DeepMind set out the results in an essay on X on November 6, writing that the agents made two significant cryptographic advances in just weeks and that such advances historically took years or decades.

Kohli also framed the lesson bluntly: in a new era of AI-enabled mathematics, we cannot treat cryptographic hardness as a fixed assumption.

FourWeekMBA reports that Pushmeet Kohli's essay gives three headline gains: a theoretical roughly 1,600x improvement for signature forgery with a short 32-byte fingerprint, about 30x lower average cost per additional key when factoring in batches, and a 3.5x CPU-time saving when the padding attack was reproduced on an already-solved 1024-bit benchmark key in a sandboxed simulation.

The essay says the 1,600x figure is a theoretical estimate and that the only result run in practice was on a key already solved by an earlier academic study. FourWeekMBA also notes the technical papers were not found on arXiv or the IACR ePrint archive, and no independent assessment was located.

Both pack sources stress the same limit: the findings target obsolete 1024-bit RSA systems, and modern protocols using 2048-bit or larger keys remain secure against this specific threat.

Crypto Briefing notes the research arrives alongside other 2026 findings involving attacks on post-quantum candidates such as HAWK, and DeepMind argues AI can help defenders improve security standards well ahead of future threats.