DeepMind's AI agents find cheaper routes to attacking 1024-bit RSA keys
The 2048-bit keys used in modern protocols are still considered secure, according to the company.
Google DeepMind says AI agents working under human supervision found two ways to make attacks on 1024-bit RSA keys cheaper, though the company stresses that the 2048-bit and larger keys used in modern protocols remain secure. One result shows that when an attacker controls roughly half of a signed block or more, the padding in the PKCS#1 v1.5 format offers no protection, and an earlier academic forgery attack was re-run on a 1024-bit key with CPU time cut 3.5-fold. The second result is an algorithm that could make factoring each additional key in a batch of same-size keys about 30 times cheaper on average, but that figure is theoretical and the method has not been implemented. Separately, DeepMind used agentic AI to formally verify key cryptographic guarantees of the Signal protocol in four weeks, finding no new vulnerabilities in the portions it examined while correctly flagging known old bugs in earlier Signal versions. The analysis did not cover the whole Signal app, and the researchers describe it as one milestone toward scaling formal verification, with its cost and any commercial offering left unclear.
Why it matters
The results mainly matter for systems that still rely on 1024-bit RSA with PKCS#1 v1.5 padding, where the padding gives no protection once an attacker controls about half of a signed block. The factoring estimate of roughly 30 times cheaper per additional key is theoretical and has not been implemented, so it is not yet a measured change in attack cost. The Signal result shows formal verification can be run on a messaging protocol in weeks, but it covers only part of the protocol and not the full app, so readers should not treat it as a broad security certification.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.