AI found two flaws in RSA-1024 encryption: what it means for your business (and what it does not)
Direct answer
On November 6, 2026, Google DeepMind announced that AI agents, guided by human cryptographers, found two weaknesses in 1024-bit RSA encryption, a standard that standards bodies retired for new signatures more than a decade ago. One weakness makes it easier to forge digital signatures when the attacker controls part of the signed data; the other makes it cheaper to factor many 1024-bit keys at once. DeepMind is clear on the point that matters: modern 2048-bit and larger keys are unaffected. This is a research demonstration on obsolete systems, with AI agents as an accelerator of scientific work, not a security event that breaks current encryption.
What the agents found, in numbers
- Signature forgery with 3.5 times less CPU time: the agents repeated a 1024-bit RSA signature attack published by academics in September 2026, using 3.5 times less CPU time than the original study. The saving came from implementation and execution, according to the paper itself, not from a new algorithm.
- A theoretical estimate 1,600 times better: for 32-byte fingerprints, the new padding-attack algorithm improves on the best prior method by roughly 1,600 times, in theoretical calculation. It is an estimate, not a measured attack time.
- Batch factoring, about 30 times cheaper: with a one-time precomputation equal to the cost of factoring one key, each additional 1024-bit key would cost about 1/58 of a full run, roughly 30 times less than the best previous batch method. The authors warn this is a leading-order estimate and the method is not implemented.
- The work, measured: two language-model agents wrote about 33,500 lines of new code, ran 336 distributed experiments and made 16,292 tool calls in 3 days and 22 hours. The human researcher sent about 20 messages, almost all approvals.
- Human-guided, not autonomous: DeepMind presents the project as AI-assisted research, with cryptographers in charge. The authors include DeepMind researchers alongside Nadia Heninger (UC San Diego) and Emmanuel Thomé (Inria).
What your business should do with this news
- Check whether you still run RSA-1024 anywhere: old certificates, embedded hardware, legacy systems and old signatures. The standard has not been allowed for new signatures since the end of 2013, but it survives in old equipment. That inventory is your risk, not the DeepMind announcement.
- AI already accelerates security research: what a human team took five months to compute, the agents repeated in under four days. Defenders and attackers get the same acceleration. Your patching cycle has to run at that speed.
- Separate announcement from availability: the papers are working papers, not yet peer reviewed, and one of the methods is not implemented. A theoretical 1,600-times figure is not an attack that exists today.
- The post-quantum transition keeps moving: standards are migrating toward cryptography resistant to quantum computing. If your company is still debating whether to migrate, this announcement is a reminder that old encryption degrades over time, with or without AI.
Note: this article summarizes Google DeepMind's November 6, 2026 announcement, the essay by Pushmeet Kohli, DeepMind's vice president of science, and the two working papers published November 5 and 6 ("Modular Roots from a Restricted Oracle" and "Fan Factories: Multiple Number Fields in Factorization Factories and Batch NFS"), with coverage from specialist media. The 1,600-times and 30-times figures are the authors' theoretical estimates, not measured results and not independently replicated.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- OpenAI launched GPT-6 Cyber: the AI that hunts vulnerabilities
- Trump signed an AI safety pact with no fines and no teeth: what changes for your company
- Privacy and data: what happens to your customers' information
Frequently asked questions
Did DeepMind's AI break RSA encryption?
No. The agents found two weaknesses in 1024-bit RSA, a key size retired more than a decade ago. Modern 2048-bit and larger keys are unaffected, according to DeepMind and the papers themselves.
What is RSA-1024 and why is it no longer used?
RSA-1024 is a 1024-bit encryption key, the digital-signature standard of years ago. Standards bodies have not allowed keys shorter than 2048 bits for new signatures since the end of 2013, but 1024-bit keys survive in old hardware and systems.
How much faster did the AI agents work?
They repeated an academic signature-forgery attack with 3.5 times less CPU time than the original study, in 3 days and 22 hours, with about 20 human messages. The saving was implementation, not a new algorithm.
What should I check in my business?
An inventory of where RSA-1024 still runs: old certificates, embedded equipment and legacy systems. That inventory is the real risk, and migrating to modern keys and post-quantum cryptography is the work to do.
