DeepMind: supervised AI agents find two weaknesses in 1024-bit RSA

Google DeepMind announced on November 6, 2026 that AI agents operating under human supervision had identified two weaknesses in legacy 1024-bit RSA cryptography. The findings are set out in two working papers published on November 5 and November 6, 2026, titled Modular Roots from a Restricted Oracle and Fan Factories, neither of which has been peer-reviewed. According to the reporting, the theoretical estimate for the signature-forgery direction is roughly 1,600 times better than the previous best-known approach. Materials reviewed here contain no detail on whether the papers have been accepted anywhere, who the co-authors are beyond the partial listing of Krishnamurthy (Dj), or what implications the weaknesses may have for real-world deployments of legacy 1024-bit RSA. The two papers cover distinct lines of work: the first treats modular roots accessible through a restricted oracle, while the second develops the fan-factory construction. Both are described as results from agents working under human oversight rather than fully autonomous systems. As of the latest available report, no follow-up has been published, and the materials do not address whether the findings have been shared with standardization bodies or cryptographic research communities.