On November 6, 2026, Google DeepMind announced that its AI agents, supervised by human experts, had found two new weaknesses in obsolete 1024-bit RSA encryption. The news was shared by Pushmeet Kohli, DeepMind's vice president of science; the work was led by Dj Dvijotham and Abhradeep Thakurta, together with Nadia Heninger of UC San Diego and Emmanuel Thomé of Inria, and two technical papers have been published. The conclusion up front: only 1024-bit RSA is affected. Modern systems overwhelmingly use 2048-bit keys or larger, which are untouched, and online banking, messaging apps, software updates and digital currencies are not threatened.
What the two weaknesses are
The first targets digital signatures. The agents found that once an attacker controls half or more of the signed block, padding adds almost no extra protection, making forgery about as easy as if there were no padding at all. On a solved 1024-bit benchmark key from earlier academic work, the agents' implementation reproduced the forgery with 3.5 times less CPU time, using under a thousandth of the computing power needed to factor the key itself. In the harder case, where the attacker controls only a small slice of the block, such as a 32-byte digital fingerprint, the agents produced a new algorithm that, on DeepMind's theoretical estimates, improves on the best previously known method by roughly 1,600 times for 1024-bit keys.
The second targets batch factoring. The agents found a new algorithm with better asymptotic complexity, cutting the average cost of factoring each additional 1024-bit key by about 30 times compared with previous batch methods. Both experiments ran in a sandbox against 1024-bit RSA, a standard the US National Institute of Standards and Technology stopped accepting for generating new signatures back in 2013. The researchers chose it precisely because it is complex enough to be meaningful and retired enough that attacking it harms nobody.
How to read the numbers, and who is affected
The 1,600x and 30x figures need to be read separately. The 1,600x figure is a theoretical estimate that applies only where the attacker controls a small part of the block; the measured number is the 3.5x CPU saving. The 30x figure is likewise an asymptotic algorithmic improvement, not a batch of real keys already broken. In the same spirit of letting AI hunt for flaws, OpenAI's GPT-6 Cyber focuses on automatically finding software vulnerabilities, with its privilege-escalation rate down to 0.07%.
What really matters here is not that "RSA has been broken" but that AI agents can now advance algorithms on their own in a hard mathematical field like cryptanalysis. DeepMind is explicit that 2048-bit RSA and NIST-standardized post-quantum algorithms are unaffected by these results. For organizations still running legacy 1024-bit RSA systems, the real to-do item remains migration, not panic: 1024-bit should have been retired long ago. What comes next is whether the two papers survive scrutiny from the cryptography community, and which algorithm agent-led cryptanalysis turns to next.
