IBM Research and cloud provider CoreWeave announced joint engineering of identity and workload controls for AI agents. IBM built a large H100 GPU cluster with CoreWeave’s infrastructure and added CoreWeave Sandboxes for isolated execution, either on dedicated hardware or a managed serverless runtime. The controls let researchers choose where agent code runs and what resources it can access, while IBM measures the security‑performance trade‑offs.
Why it matters
Researchers can test reinforcement‑learning agents in secure, isolated environments, lowering the risk of data or network breaches.
