# Google DeepMind AI research in cryptography

> Developing situation record from CLSTR: https://clstr.news/situations/google-deepmind-ai-research-in-cryptography
> Updated: 2026-11-10T12:00:00.000Z. Sources: 16. Developments: 2.

Google DeepMind has utilized AI agents to both identify vulnerabilities in legacy encryption and verify the security of modern protocols.

In early November 2026, researchers announced that AI agents discovered two significant weaknesses in 1024-bit RSA cryptography: digital signature forgery and batch factorization. The findings suggest that signature forgery could be 1,600 times more effective than previous methods, while batch factorization could reduce the total cost of factoring to approximately 1/58th of the cost of the general number field sieve method.

Following these discoveries, researchers applied a new AI agentic framework, AlphaProof Code, to formally verify key components of the Signal protocol. In a four-week project, the agents generated approximately 6.5 million lines of Lean code to verify the post-quantum PQXDH handshake and the Triple Ratchet algorithm. While the study did not uncover new vulnerabilities in the analyzed sections, it successfully identified previously patched flaws in older versions of the protocol. The research demonstrated that verification could extend to actual compiled machine code, offering potential applications for securing medical device firmware and power grid software.

## Timeline

### 2026-11-10: Google DeepMind uses AI to mathematically verify Signal protocol

Google DeepMind researchers used the AlphaProof Code AI framework to mathematically prove the security of Signal protocol's core cryptographic implementations in just four weeks.

11 sources. https://clstr.news/cluster/google-deepmind-uses-ai-to-formally-verify-signal-messaging-protocol

### 2026-11-07: Google DeepMind AI agents identify weaknesses in legacy RSA-1024 cryptography

Google DeepMind AI agents discovered two vulnerabilities in legacy 1024-bit RSA cryptography, significantly improving the efficiency of digital signature forgery and batch factorization.

5 sources. https://clstr.news/cluster/google-deepmind-ai-agents-identify-weaknesses-in-legacy-rsa-1024-cryptography

---
Cite as: Google DeepMind AI research in cryptography. CLSTR, https://clstr.news/situations/google-deepmind-ai-research-in-cryptography
