DeepMind's AI agents find new weaknesses in outdated 1024-bit RSA encryption

Google DeepMind announced on November 6 that AI agents, guided by human researchers, uncovered two significant weaknesses in legacy 1024-bit RSA encryption, which is already considered outdated and insecure for serious use. The first weakness involves padded digital signatures: once an attacker controls about half of a signed block, the padding meant to add protection stops being effective. In a controlled test using an already-broken 1024-bit benchmark key, the agents reproduced a signature forgery using 3.5 times less CPU time, and under a thousandth of the computing power normally needed to break the key outright. For short, 32-byte fingerprints specifically, the team's new algorithm theoretically improves on the best previous method by about 1,600 times. The second weakness is a new batch factoring algorithm that cuts the average cost of cracking each additional 1024-bit key by roughly 30 times compared to older batch techniques. DeepMind was clear that this only affects the already-obsolete 1024-bit RSA standard, and that modern 2048-bit keys and larger remain secure. The research was led by Dj Dvijotham and Abhradeep Thakurta, working with outside academic researchers.

Why it matters

Even though only obsolete 1024-bit keys are affected, it shows AI agents can meaningfully accelerate cryptographic weakness research.

Source: Google DeepMind