Google DeepMind: human-guided AI agents find two new weaknesses in legacy 1024-bit RSA
Confirmed
Importance: major (4 of 5)The takeaway
No key was broken in practice and DeepMind says 2048-bit RSA is unaffected.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 8 November 2026
Your AI and this story
- GPT-6 Astra190 days after its cutoff
- Claude Opus 5.5129 days after its cutoff
- Gemini 4 Argon220 days after its cutoff
- Grok 4.7159 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 129 days before it.
Key facts
- Scope stated by DeepMind: ‘these results only affect obsolete 1024-bit RSA systems’; modern protocols using 2048-bit or larger keys ‘remain secure’
- Signatures: once an attacker controls roughly half or more of the signed block, padding adds no protection against forgery (chosen inputs to a signing device)
- Demonstration in a sandbox on an already-factored 1024-bit benchmark key from an earlier academic study: forgery reproduced with 3.5x less CPU time than the prior method, under 1/1000 of the cost of factoring the key
- For short controlled inputs (e.g. a 32-byte hash) a new algorithm improves on the previous best by ~1,600x for 1024-bit keys — a theoretical estimate, not run
- Batch factoring: new algorithm with better asymptotic complexity (‘factorization factories’, batch NFS), ~30x lower average cost per additional 1024-bit key versus previous batch methods
- Led by Dj Dvijotham and Abhradeep Thakurta (DeepMind Science), with Nadia Heninger (UC San Diego, Inria) and Emmanuel Thomé (Inria); announced by VP Science Pushmeet Kohli in an X Article
- Papers are PDFs on DeepMind’s storage, not peer-reviewed at announcement; the role split between agents and humans is described only as ‘guided by humans’
What happened
Google DeepMind’s Science team announced on Nov 6, 2026 (X post and an X Article by Pushmeet Kohli) that AI agents “guided by human supervision” uncovered “two major weaknesses on legacy 1024-bit RSA-based cryptosystems, showing that these systems can be compromised with a lower computational budget than previously thought possible.” Details are in two technical papers linked from the article. Viral summaries (e.g. “Google breaks RSA keys”) overstate it: no new key was factored; the forgery demo used a key already factored in earlier work, and the 1,600x and 30x figures are cost estimates.
Why it matters
Another lab result (after Anthropic’s July cryptanalysis work) arguing that AI agents can shift cryptographic hardness estimates faster than the usual years-long cadence; DeepMind frames it as a reason for proactive stress-testing.
Sources
4 sources from 2 sites. Numbers match the chips in the text.
4 sources: 4 primary
Primary
- Pushmeet Kohli (X Article): Advancing the frontier of cryptanalysisx.com, official
- Google DeepMind on X (136k views)x.com, official
- Paper: Faster affine root and a 1024-bit computation (signatures)storage.googleapis.com, paper
- Paper: Multiple number fields in factorization factories and batch NFSstorage.googleapis.com, paper
Changes
- Filed (missed on 11-07)