Google DeepMind announced on 6 November 2026 that AI agents, working under human direction, had found two weaknesses in legacy 1024-bit RSA encryption. Headlines about "AI breaking encryption" followed within hours. The real story is more interesting — and much less alarming — than those headlines suggest.
What DeepMind actually did
The work was published as two working papers on 5 and 6 November, titled "Modular Roots from a Restricted Oracle" and "Fan Factories". It was led by DeepMind researchers Dj Dvijotham and Abhradeep Thakurta, working with cryptographers Nadia Heninger of UC San Diego and Emmanuel Thomé of Inria. The team built on a September 2026 preprint from UC San Diego and Inria that demonstrated forgery of 1024-bit RSA signatures.
Crucially, the experiments ran in an isolated sandbox. No live systems and no live keys were touched. And the papers have not yet been peer-reviewed, so the wider cryptography community has not yet had its chance to check the work. The figures below are as reported by secondary outlets covering the announcement.
Weakness one: forging signatures more cheaply
The first result concerns digital signatures — the mechanism that proves a message or software update really came from who it claims to come from. The researchers found a new way to forge signatures on 1024-bit RSA keys when the attacker controls roughly half or more of the signed data. In that situation, the padding that is supposed to protect the signature offers little meaningful protection.
In a measured rerun against an already-solved 1024-bit benchmark key, the agents' implementation reproduced the forgery using about 3.5 times less computing time than the previous approach — 391 CPU core-years against 1,380 for the original team. For a harder case, where the attacker controls only a short 32-byte fingerprint, a new algorithm improves on the previous best method by about 1,600 times. That 1,600 figure is a theoretical estimate from the maths, not a measured experimental speedup, and it is important not to read it as "1,600 times faster in practice".
Weakness two: cracking many keys at once
The second result is about batch factoring — breaking many RSA keys together rather than one at a time. The researchers describe a new algorithm, nicknamed a "fan factory", that reduces the average cost of factoring each additional 1024-bit key by roughly 30 times compared with earlier batch methods. Again, this is a theoretical result: nobody has built such a factory, and the improvement applies to the cost per extra key after the first, not to breaking a single key from scratch.
What is NOT affected
This is the part that got lost in many headlines. The results apply only to obsolete 1024-bit RSA. DeepMind itself says the findings do not threaten modern deployments using 2048-bit keys or larger. Online banking, messaging apps, software updates and digital currencies are not threatened by these specific findings.
That is because 1024-bit RSA has been retired for years. NIST, the US standards body, stopped accepting 1024-bit RSA for generating new signatures back in 2013. Any system still using 1024-bit keys today is running on borrowed time for reasons that have nothing to do with AI — it was already considered too weak, and security guidance has said so for over a decade.
Why this still matters
If nothing you use is affected, why should you care? Three reasons. First, it shows that AI agents are becoming genuine research tools in mathematics and cryptography, not just chatbots. The agents did exploratory work, found and fixed problems in the algorithm, and produced results that human specialists then checked and developed. That is a new way of doing science, and it will show up in other fields next.
Second, it is a warning for the long tail of old systems. Somewhere out there, old devices, legacy software and forgotten servers still use 1024-bit keys. If you manage systems, this is a good moment to check for them and retire them. The 1024-bit era is over; the only question is whether your systems got the message.
Third, it is a reminder to read past the headline. "AI cracks encryption" is a frightening sentence. "AI agents help researchers find weaknesses in a retired key size, while modern encryption is unaffected" is the same story, told accurately. In security news, the difference between those two sentences is everything.
Key takeaways
- DeepMind reported on 6 November that AI agents, guided by humans, found two weaknesses in legacy 1024-bit RSA.
- Signature forgery was reproduced with about 3.5 times less computing time; a harder case is about 1,600 times better in theory.
- A new batch-factoring method cuts the average cost per extra 1024-bit key by about 30 times, in theory.
- Modern 2048-bit RSA and above are not affected, according to DeepMind.
- NIST stopped accepting 1024-bit RSA for new signatures in 2013 — systems still using it should retire it.
Conclusion
DeepMind's RSA result is a genuine advance in cryptanalysis, and a clear sign that AI agents can now contribute to hard mathematical research. It is not the end of encryption. Your bank, your messages and your software updates all rely on key sizes this research does not touch. The real lesson is narrower and more useful: retire old 1024-bit keys wherever they still exist, and treat every "AI breaks encryption" headline with the same care you would give any security claim — read what was actually tested, and on what.
FAQ
Has AI broken RSA encryption?
No. The research affects only legacy 1024-bit RSA, a key size retired years ago. DeepMind says modern 2048-bit RSA and above are not affected by these findings.
What does "1,600 times" actually mean?
It is a theoretical estimate for one specific, hard forgery case — not a measured speedup from a real attack. The measured result was a rerun of a known forgery using about 3.5 times less computing time.
Should I change my passwords because of this?
No. This research has nothing to do with passwords. It concerns old RSA keys used in some digital signatures and encryption. If you run servers or devices, check that none of them still use 1024-bit RSA keys.
Sources
- Mundo IA: Google DeepMind uses AI agents to investigate the security of cryptographic systems — https://en.mundoia.com.ar/google-deepmind-ia-vulnerabilidades-rsa-1024-bits/
- Vandata: DeepMind RSA-1024 Weaknesses: What Data Teams Should Fix — https://vandatateam.com/blog/deepmind-rsa-1024-weaknesses
- ExplainX: DeepMind 1024-Bit RSA Claim: What Is Actually Verified? — https://explainx.ai/blog/deepmind-ai-agents-two-weaknesses-1024-bit-rsa-claimed-vs-verified-2026
Related reading on Daily Tech & AI:
- GPT-6 Cyber: OpenAI's New Security AI Explained — https://amjadphd.blogspot.com/2026/11/gpt-6-cyber-openai-security-model.html
- AI Is Speeding Up AI Research: METR Finds 1.9x Faster — https://amjadphd.blogspot.com/2026/11/metr-ai-research-acceleration.html
- 5 Always-On AI Agents Launched This Week — https://amjadphd.blogspot.com/2026/11/always-on-ai-agents-week.html
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for one tech and AI update every day: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r

Comments
Post a Comment