OpenAI Private Intelligence: what it actually gives businesses

On Friday morning, October 30, OpenAI announced OpenAI Private Intelligence on X: "Privacy should advance as fast as intelligence." The new product page gathers OpenAI's business privacy protections under one name. It's less a single new product than a bundle, and one part of it, Private Inference, is new and still in preview.

What's in it. OpenAI's page lists four promises and three mechanisms:

• Access restrictions. OpenAI says it sets "strict limits" on its own access to business content, and offers Zero Data Retention (ZDR) and Private Inference "for eligible workflows."
• Data ownership. You own your business data, including inputs and outputs.
• Agent actions. Admins decide what agents can access, which tools and actions they can use, and when they need approval.
• Data security. Encryption, access controls and threat detection.
• Zero Data Retention. For eligible API customers, OpenAI doesn't keep prompts or model responses after a request is processed. Since September 22 it pairs this with Private Safety Processing (PSP): flagged content is stored encrypted in storage the customer controls, reviewed only by an automated system inside OpenAI, and OpenAI employees can't get the decryption keys. OpenAI says the retained content is used only for safety and never for training.
• Private Inference (preview). An API option that combines confidential computing with "strict, verifiable controls" to protect prompts and responses while the model processes them.
• Compliance. SOC 2 Type 2, ISO certifications, HIPAA support through a Business Associate Agreement for eligible services, and FedRAMP support for designated government offerings.

Who it's for. Businesses handling sensitive data: OpenAI's earlier post names financial records, health data, confidential plans and proprietary research. A second page covers controls across ChatGPT, Codex and the API, including SSO and role-based access in ChatGPT for enterprises. The ZDR and Private Inference parts are API features. OpenAI says it built the protections with "some of our largest customers," but doesn't name them.

Price and availability. Not stated. The pages offer "Contact sales" and an interest form. OpenAI doesn't say which models or endpoints qualify, what Private Inference costs, or when the preview ends.

The caveats. "Eligible" carries a lot of weight here, and OpenAI doesn't define it publicly. Its own white paper says Private Inference "does not automatically cover the initial API request" or "every OpenAI system," and that it covers first-party OpenAI traffic only, not Amazon Bedrock or other third-party resellers. Even under ZDR, OpenAI says images flagged as potential child sexual abuse material are still kept for manual review and reporting, as the law requires.

Our take. For a small team calling the API, nothing here changes your bill or setup today. ZDR is only for eligible API customers, and Private Inference is a preview you'd ask sales about. For companies whose lawyers have kept them off frontier models, this gives them a specific checklist to take to OpenAI. MODEL has not tested Private Intelligence. This summary is based only on OpenAI's published pages and white paper, read October 30, 2026.