I noted Google DeepMind’s research on AI agents finding weaknesses in legacy RSA. The results concern 1024-bit keys, not modern systems using larger keys; the papers are still working papers and have not been peer reviewed. The post raises the broader question of what happens when AI agents stress-test the mathematics security relies on.
Google DeepMind announced on 6 November that AI agents, under human supervision, found two new weaknesses in legacy 1024-bit RSA.
The first weakness is a signature forgery method: once an attacker controls about half the signed block, padding adds almost no protection.
In a sandbox test, the agents reproduced a known forgery with 3.5 times less CPU time.
For short 32-byte fingerprints, DeepMind estimates a new algorithm about 1,600 times better than the previous best, in theory.
The second weakness is a batch factoring algorithm that cuts the average cost of each extra 1024-bit key by about 30 times.
The post says modern systems use 2048-bit keys or larger, and those are unaffected.
The working papers are not peer reviewed yet.
I asked: What happens when AI starts stress-testing the maths our security rests on?