Google DeepMind AI agents discover cryptanalysis algorithms that improve attacks on RSA-1024, including a theoretical 1,600× speedup in a signature-forgery method

Google DeepMind researchers, working with human supervision and collaborators including Nadia Heninger and Emmanuel Thomé, used AI agents to find two weaknesses in legacy 1024-bit RSA systems. They showed padding offers no protection once an attacker controls roughly half a signed block, reproducing a prior forgery result with 3.5× less CPU time, and derived a new algorithm estimated 1,600× faster for 1024-bit keys. A second algorithm cuts average batch factoring cost per key by about 30×. The authors stress modern 2048-bit-plus protocols remain secure.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• Google
• AI Agents