Jack Paul 
Google DeepMind says two AI agents carried out a known 1024-bit RSA cryptography computation over nearly four days with only about 20 human messages. Most were approvals. The work covered coding, large-scale execution, and debugging.
A sandboxed reproduction
The test involved RSA digital signatures. After billions of advance queries to a simulated signing service, the agents computed a mathematical root needed for a new signature once access ended. They did not recover a private key. DeepMind researcher Pushmeet announced the work on Nov. 6. The paper calls it a sandboxed reproduction of a previously solved benchmark instance and says it does not threaten deployed systems based on 2048-bit or larger RSA.
The human chose the problem, instance, and limits on the signing service. The person also approved conditions for irreversible steps. According to the paper, the researcher wrote no code, launched no jobs, and diagnosed no faults.
How the agents worked
A supervisor agent kept the design, budgets, and checks. An executor wrote and tested code, launched distributed jobs, and repaired failures. The supervisor also brought in a specialist sub-agent to speed one linear-algebra step. The agents produced about 33,500 new lines of code, ran 336 distributed experiments, and exchanged 218 messages. From design to verified result, the work took three days and 22 hours.
Their first computed root was wrong because a 32-bit offset overflowed. A correctness check caught the error. The agents found and fixed it without repeating the advance signing queries. Still, the limited human messaging came with a human-selected task and approval gates.
What the speedup measures
The earlier academic computation took about five months. DeepMind’s run took 3.2 days from its first full-scale sieving job to the verified root. It reached about 300,000 CPU cores and 16 GPUs at peak. The authors say the calendar-time difference mainly came from parallelism.
The CPU saving came from implementation and execution changes within the existing algorithm. The 391-core-year total excludes about 330 reserved GPU-hours, of which about 100 were busy, and 7.5 core-years of probes and rehearsals. The CPU totals also come from different hardware, so this is not a controlled like-for-like comparison.
The cryptographic setup matters too. A separate program holding the modulus factorization answered roughly 4 billion advance queries. It was withdrawn before the target was drawn. The agents then computed the target’s root without further queries.
The measured reproduction did not demonstrate the paper’s new algorithm for more restricted signing access at cryptographic size. A separate Fan Factories paper estimates roughly 30-fold lower costs per additional 1024-bit target than the multiple number field sieve, after shared precomputation costing about one general number field sieve factorization. Those are leading-order heuristic estimates. No fan factory has been implemented.
The working paper identifies the principal models only as A and B and links no public implementation repository. That limits outside comparisons.