Return to Threats

Google DeepMind’s AI Agents Make 1024-Bit RSA Keys Cheaper to Attack

securityweek.com 2026-11-09 malicious AI use High

Executive Summary

Human-supervised AI agents made attacks against legacy RSA-1024 systems cheaper, including signature forgery and theoretical batch factoring. Security, cryptography, and infrastructure teams operating legacy RSA-1024 systems should prioritize migration to stronger key sizes.

What Happened

Working with cryptographers, DeepMind’s AI agents cut the cost of forging RSA-1024 signatures and found a cheaper way to factor keys. The post Google DeepMind’s AI Agents Make 1024-Bit RSA Keys Cheaper to Attack appeared first on SecurityWeek .

Affected Products

  • RSA-1024

Vulnerability / Attack Technique

AI-assisted cryptanalysis and RSA signature forgery

Technical Mechanism

The agents reproduced an earlier RSA-1024 signature-forgery attack with lower CPU cost and found a theoretical algorithm that could reduce the average cost of factoring each additional RSA-1024 key after a substantial initial computation. The reported factoring method has not been implemented, and the signature attack does not recover the private key.

AI Security Category

AI-assisted attack

Part of the AI-Assisted Attacks topic.

Relevant Frameworks

MITRE ATLAS: Impact

RealGround Analysis

This demonstrates how AI agents can improve attack execution and cryptanalytic research against non-AI targets, even when operating under human supervision. Teams running AI agents or LLM applications should govern access to computational resources, monitor agent-directed security research and code execution, and include legacy cryptography discovery in AI-assisted attack assessments.

Recommended Controls

  • Inventory certificates, keys, signing services, and protocols to identify RSA-1024 usage.
  • Migrate RSA-1024 keys and signatures to modern cryptographic configurations using at least 2048-bit RSA or an approved stronger alternative.
  • Restrict AI-agent access to cryptographic tooling, high-performance compute, signing interfaces, and sensitive key-management systems.
  • Require human approval, sandboxing, logging, and resource limits for agent-generated cryptanalysis code and computations.
  • Conduct AI red-team exercises covering AI-assisted attacks against legacy cryptographic assets and signing workflows.

Related RealGround Service

AI Red Teaming · or start with the free AI Agent Security Scanner.

Sources

securityweek.com: https://www.securityweek.com/google-deepminds-ai-agents-make-1024-bit-rsa-keys-cheaper-to-attack/

Summary and analysis written by RealGround from the source above; last updated 2026-11-09.

Talk to AI CISO