In September 2026, researcher Stephen Weis announced the successful factorization of RSA-896, a large integer used in RSA cryptographic challenges. The computation required approximately 30 GPU-years of processing power, executed over a 10-day period using a fleet of up to 2048 GPUs. The project utilized idle compute capacity, demonstrating that large-scale factoring tasks can be performed by leveraging existing data center infrastructure. To achieve this, Weis used the AI model Claude to port the CADO-NFS (Crible Algébrique: Distribution, Optimisation - Number Field Sieve) software to run on GPUs. CADO-NFS is a widely used open-source implementation of the General Number Field Sieve (GNFS), the most efficient known algorithm for factoring large integers. Weis emphasized that the project did not introduce new algorithmic improvements to the GNFS, which remains subexponential in complexity. Consequently, the result does not pose a new threat to currently deployed RSA-2048 keys. However, the successful factorization of RSA-896 serves as a practical demonstration that RSA-1024 keys are increasingly vulnerable to entities with access to significant data center-level GPU resources. The project relied on established algorithms and codebases developed over several decades, with the AI's role primarily focused on the technical porting and orchestration of the distributed workload.