OpenAI bundles business privacy into Private Intelligence – its core is still in preview

OpenAI is bundling its privacy promises for business customers under one name: OpenAI Private Intelligence. The company launched the package on X on Friday 30 October at 15:12 UTC, arguing that privacy should advance as fast as intelligence. The new piece is Private Inference, which OpenAI itself labels an API preview.

For leaders who want to use the most capable models on sensitive data, this is OpenAI's answer to a central procurement question: who can see our data, and where does it go?

Four promises on one page

On the product page, OpenAI splits the package into four parts:

• 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 what it calls eligible workflows.
• Data ownership: Customers own their business data, including inputs and outputs.
• Agent actions: Customers decide what agents can access, which actions they can take and when they need approval.
• Data security: Encryption, strict access controls and ongoing threat detection.

OpenAI writes that it is building these protections in close collaboration with some of its largest customers. The page also points to SOC 2 Type 2 audits, ISO certifications, HIPAA support through Business Associate Agreements for eligible services and FedRAMP support for designated US government offerings.

Much of this existed before. What is new is that OpenAI packages it as one product and puts Private Inference up front.

The new part: confidential computing in preview

The product page describes Private Inference as confidential computing with strict, verifiable controls designed to protect prompts and responses while the model processes them. The diagram on the page places it in the middle of the flow between the customer's prompt and the response, tagged "API Preview".

OpenAI's own technical white paper on Private Safety Processing describes the scope more narrowly. It says Private Inference is an optional protection that extends operator-access protections and hardware attestation to eligible safety-inference workloads. According to the white paper, it does not automatically cover the initial API request, nor every OpenAI system. OpenAI writes that it expects to extend confidential-computing protection beyond the safety stack when the necessary technology, capacity and product commitments are met, but gives no date.

The white paper also says Private Inference is only available to customers that meet eligibility and capacity requirements. The sources do not say what those requirements are.

Zero retention, but the customer stores for 30 days

The second pillar is Zero Data Retention with Private Safety Processing. According to OpenAI's developer documentation, OpenAI does not retain customer prompts and responses. Content selected for automated safety review is encrypted and written to storage the customer controls. It can only be decrypted in a hardware-attested environment that disables human access, and it is not used to train models.

The arrangement also puts obligations on the customer. The documentation requires customers to register and validate their own storage, retain the encrypted records for at least 30 days, keep key and storage permissions in order and respond when OpenAI raises safety concerns.

What this means for leaders

• Ask about eligibility, not just the name. Both ZDR and Private Inference apply to "eligible" workflows. Get a written answer on which models, endpoints and services are actually covered before sensitive data goes in.
• Treat the preview as a preview. OpenAI states neither price, which models are covered nor when Private Inference becomes generally available. Per the white paper, the protection today covers safety review, not the full model call.
• Agent governance is a leadership decision. When the customer sets the limits for what agents may do and when they must ask, someone in the organisation has to own those rules.
• Thirty days of storage on your side is an operational task. Storage, keys and permissions must be set up and maintained, and that belongs in the risk assessment.

OpenAI offers an interest form for organisations that want to learn more. The company has not stated pricing or availability in Europe in the sources hogby.ai reviewed.

Sources and media

• OpenAI on X: announcement of Private Intelligence, 30 October 2026, 15:12 UTC
• OpenAI Private Intelligence (product page)
• OpenAI: Register interest in Private Intelligence
• OpenAI documentation: ZDR with Private Safety Processing
• OpenAI: Private Safety Processing Overview (white paper, PDF)

Illustration: AI-generated with Grok Bot / hogby.ai