Google Cloud Expands Confidential Computing to Strengthen AI Inference Security

Confidential computing uses trusted execution environments, or TEEs, and hardware keys to close a security gap that can leave data exposed while in use. For enterprise AI, prompts, model parameters and data shared across organizations often contain trade secrets. Google Cloud is therefore extending protection beyond data storage and transmission to inference and fine-tuning, reducing the risk that cloud providers or highly privileged personnel can access plaintext data.

Google Cloud announced on June 23, 2026, that its G4 Confidential VMs had entered global preview. The VMs use NVIDIA RTX PRO 6000 Blackwell GPUs and fifth-generation AMD EPYC processors, with AMD SEV encrypting data exchanged between CPUs and GPUs. Google Cloud also open-sourced its Prompt Encryption SDKs on GitHub to provide end-to-end protection for prompts and responses. No financial amount was disclosed.