Google Cloud Adds OpenAI Models, Promises Business Data Won’t Train AI

What happened

Google Cloud announced a private preview that makes OpenAI’s GPT‑6 family available via its Marketplace. Customers can purchase usage on a single invoice, switch between Google and OpenAI models, and are told that all prompts, context, and outputs will remain within defined data boundaries and will not be fed back into model training. Google also previewed future zero‑data‑retention and Confidential Compute options.

Why it matters for trust and compliance

• The data‑boundary promise maps directly to the control objective of restricting AI model training data, a requirement for continuous‑control‑assurance programs that need verifiable evidence of AI governance.
• Demonstrate data‑boundary controls for AI model usage to satisfy AI‑governance requirements across frameworks.
• Collect continuous evidence (API logs, retention settings, Confidential Compute attestations) for audit‑ready documentation.

Who is affected

Recommended actions

1. Map your AI data‑handling policy to the VCF control objective for model training data isolation.
2. Enable detailed logging of all model API calls and verify zero‑retention configurations are active.