Google DeepMind AI Agents Accelerate Cryptographic Vulnerability Research
Google DeepMind, working with academic cryptographers, deployed supervised AI agents to discover cheaper methods for attacking legacy 1024-bit RSA systems, reducing the CPU cost of a known forgery attack by 3.5-fold and finding algorithms significantly more efficient for key factoring. For defenders, this research provides empirical, quantified evidence that closes a long-standing ambiguity about the residual risk posed by legacy RSA-1024 deployments — directly informing cryptographic migration prioritisation. The capability does not yet extend its analysis to modern key sizes, and organisations will need cryptographic inventory maturity to translate these findings into actionable migration roadmaps.