Google DeepMind working paper, developed with AI agents, lowers the theoretical cost of factoring many 1024-bit numbers

Google DeepMind · Research preprint

Read the source (opens in a new tab)
What happened
Google DeepMind researchers, some also affiliated with UC San Diego, Inria and Stanford, released "Fan Factories," a working paper dated November 5, 2026, on factoring many integers of one size more cheaply.
Key points
  • At leading order, "a precomputation of one factorization lets each further 1024-bit integer be factored at about 1/58 of the cost of factoring it with the NFS, 30 times less than with MNFS."
  • "The results were obtained with a two-agent AI harness: one model generated the proofs, and a second model verified them," with key identities machine-checked in Lean 4.
  • The authors say the results "do not threaten deployed cryptosystems today that are based on 2048-bit and larger RSA" and lower only the cost of each further integer.
  • They say the main significance is that the results "sharpen our picture of how quickly AI-assisted research can move a field that usually advances slowly."
Evidence
A theoretical working paper. Its cost figures are leading-order estimates from heuristic exponents, not measurements; the authors say no fan factory has been implemented.
Uncertain
Not peer reviewed. The full proofs and Lean source are in a "full version" the paper cites but this entry did not read. The paper does not describe how researchers directed the agents.

See it in the latest record