Humanity Unfolds

DeepMind uses AI agents to stress-test legacy RSA

Nov 6, 2026

Google DeepMind reports two advances in analyzing legacy 1024-bit RSA: one studies older signature schemes and the other estimates the cost of factoring batches of keys.

Human-guided agents reproduced a signature benchmark with about 3.5 times less CPU time. The batch-factoring paper presents a theoretical improvement under standard assumptions. The results leave deployed systems based on 2048-bit or larger RSA unaffected by these specific methods. The linked papers list the full research team.

Explore on Humanity Unfolds

People

Sources