DeepMind reports AI-assisted advances against legacy RSA
DeepMind reports human-guided research on legacy RSA; technical performance claims still need review.
An AI-assisted cryptography result needs a carefully drawn boundary around what it demonstrates.
What happened
Google DeepMind reported on 6 November that human-supervised AI agents helped develop faster methods involving legacy 1024-bit RSA and batch factoring. The company said the specific findings do not affect modern 2048-bit RSA.
Why it matters
The research points to a possible role for agents in mathematical security work. Its significance depends on the exact assumptions and attack settings, rather than a general claim that encryption has been broken.
The bigger picture
Review needed: the technical papers were located but could not be inspected during verification. The evidence is not verified enough to publish speedup multipliers, independently validate the methods or claim a practical break of modern RSA. Any draft should retain those limits rather than turn a research announcement into a universal security claim.