The short answer
● What happened — DeepMind agents found two advances against 1024-bit RSA
● Signature forgery — roughly 1,600x faster than the previous best method
● Batch factoring — about 30x cheaper per additional key
● What is affected — legacy 1024-bit RSA only
● What is not — 2048-bit keys, which is almost everything in use today
Google DeepMind published two cryptographic results on 6 November, found by AI agents working under human supervision alongside named academic cryptanalysts.
The two results
| Advance |
What the agents found |
| Signature forgery |
Padding protects less than assumed once an attacker controls about half the signed block. For short inputs such as a 32-byte fingerprint, a new algorithm estimated around 1,600x faster than the previous best for 1024-bit keys. |
| Batch factoring |
A new algorithm for factoring many same-size keys at once, lowering the average cost per additional 1024-bit key by roughly 30x. |
What this does not do
Modern encryption almost universally uses 2048-bit keys or larger, and is unaffected. DeepMind is explicit that the results apply only to obsolete 1024-bit RSA. No production system has been shown at risk, no live key was broken, and the work ran in a sandbox against an already-solved benchmark key from an earlier study.
The 1,600x figure is a theoretical estimate, not a measured attack on anything real. The reproduction that was actually run used 3.5 times less CPU time than the prior method, and less than a thousandth of the compute needed to break the underlying key.
Why it still matters
The interesting claim is not that 1024-bit RSA is weak. That has been known for years, which is why it was deprecated. The claim is about speed of discovery: agents under human supervision produced two genuine cryptanalytic advances in weeks.
If that generalises, the shelf life of every deprecated-but-still-deployed algorithm gets shorter. Anything still running 1024-bit RSA was already overdue for replacement; this is a reason to stop postponing the audit.
Who did the work
The research was led by Dj Dvijotham and Abhradeep Thakurta, with Nadia Heninger of UC San Diego and Inria and Emmanuel Thomé of Inria. The agents operated under human supervision throughout, which DeepMind states rather than downplays.
Frequently asked questions
Is my data at risk?
Not from this. Modern encryption uses 2048-bit keys or larger and is unaffected.
Does this break Bitcoin?
No. The results are specific to legacy 1024-bit RSA and were demonstrated in a sandbox against a benchmark key already solved in earlier work.
Was a real key broken?
No. The agents reproduced a known result more efficiently. The headline speed figure is an estimate.
Should I do anything?
If any system you run still uses 1024-bit RSA, replace it. That was true before this research and is more urgent now.
Did AI do this on its own?
No. DeepMind describes the agents as working under human supervision, with established cryptanalysts on the team.
Sources: Office Chai · Crypto Briefing.
Published 7 November 2026. Figures checked against the reporting linked above on the same day.