Google DeepMind Agents Cut CPU Use 3.5 Times on Known 1024-Bit RSA Benchmark

TLDR

Google DeepMind says two AI agents reproduced a known 1024-bit RSA cryptography computation in three days and 22 hours.
The human researcher sent about 20 messages, mostly approvals, and wrote no code.
The agents wrote about 33,500 lines of code and ran 336 distributed experiments.
The run used 391 CPU core-years, compared with 1,380 for an earlier team on the same instance.
The papers say the results do not threaten 2048-bit or larger RSA systems.

A researcher at Google DeepMind picked a math problem, set the rules and then mostly stepped aside. Over nearly four days, two AI agents handled the coding, the computing and the debugging, according to a working paper from the company.
The human sent about 20 messages during the run, and most of them were approvals. The paper says the researcher wrote no code, launched no jobs and diagnosed no faults.
The task was a known 1024-bit cryptography computation tied to RSA digital signatures. It was a sandboxed reproduction of a benchmark instance that another team had already solved.
DeepMind researcher Pushmeet Kohli announced the work on Nov. 6. The papers state that the results do not threaten deployed systems that use 2048-bit or larger RSA keys.

https://t.co/rROvupyPAL
— Pushmeet Kohli (@pushmeet) November 6, 2026

How the Work Was Split
The researcher chose the problem, the instance and the limits on a simulated signing service. The researcher also approved the conditions for steps that could not be undone.
A supervisor agent kept track of the design, budgets and checks. An executor agent wrote and tested code, launched distributed jobs and fixed failures.
The supervisor also brought in a specialist sub-agent to speed up one linear algebra step. Across the run, the agents exchanged 218 messages and ran 336 distributed experiments.
The agents produced about 33,500 new lines of code. From design to the verified result, the work took three days and 22 hours.
The setup had several stages. A separate program that held the modulus factorization answered roughly 4 billion advance queries, then was withdrawn before the target was chosen.
The agents then computed the mathematical root needed for a new signature without any more queries. They did not recover a private key.
A Bug Caught Along the Way
The first root the agents computed was wrong. The cause was a 32-bit offset that overflowed.
A correctness check flagged the error. The agents found and fixed it without repeating the advance signing queries.
The completed computation used 391 CPU core-years, according to the paper. An earlier team needed a rounded 1,380 core-years on the same instance and algorithm, which works out to about 3.5 times less CPU use.
That figure excludes about 330 reserved GPU-hours, of which about 100 were busy, plus 7.5 core-years of probes and rehearsals. The two totals also come from different hardware, so the comparison is not like-for-like.
The earlier academic effort took about five months. DeepMind’s run took 3.2 days from its first full-scale sieving job to the verified root, peaking at about 300,000 CPU cores and 16 GPUs, and the authors credit parallelism for most of that gap.
The run did not test the paper’s new algorithm for more restricted signing access at full cryptographic size. A separate Fan Factories paper estimates about 30-fold lower costs per extra 1024-bit target after shared precomputation, but no fan factory has been built.
The working paper names the main models only as A and B. It does not link to a public code repository, which limits outside comparison.
The post Google DeepMind Agents Cut CPU Use 3.5 Times on Known 1024-Bit RSA Benchmark appeared first on Blockonomi.