DeepMind says agents found two weaknesses in legacy 1024-bit RSA
On 6 November 2026 Pushmeet Kohli said human-guided agents found two weaknesses in legacy 1024-bit RSA. He wrote that RSA 2048 is not affected. This desk did not open the papers.
On 6 November 2026, Pushmeet Kohli, VP at DeepMind, posted that the Google DeepMind science team had used AI agents, with human supervision, to find two weaknesses in legacy 1024-bit RSA systems. His essay says those systems can be compromised with a lower computational budget than previously thought possible.
He names the two results without the method. One can bypass older digital signature protections. The other factors legacy 1024-bit RSA keys in bulk. He writes that the results do not affect modern schemes such as RSA 2048. He also writes that the work is a reason not to treat the old mathematical hardness of those schemes as settled, and that agents can help defenders retest standards.
The short post and the essay went up the same afternoon. When this desk read them, the short post showed 49,100 views and the essay showed 124,500. His X bio lists him as Chief Scientist at Google Cloud and VP at DeepMind. The essay links two technical papers, one on digital signatures and one on key factorisation. This desk did not open either PDF.
What stays out
A security note from this desk carries the name, the access limit, and what the source says still stands. It does not carry the attack. There is no speedup factor here, no block size, and no description of a method. Those details, if they are in the papers, were not read.
Kohli’s own boundary is the one in the post: RSA 2048 is not affected by these results. The posts do not say how many 1024-bit keys are still in use.