OpenAI unveils Private Intelligence to bolster enterprise AI privacy

At DevDay 2026, OpenAI announced Private Intelligence, a privacy initiative aimed at strengthening its zero-data-retention infrastructure for enterprise AI customers. CEO Sam Altman showcased the two main components during the keynote: ZDR with Private Safety Processing (ZDR with PSP), available now, and Private Inference, expected to launch in fall 2026.

ZDR with PSP is designed to perform automated safety reviews without storing customer content, allowing OpenAI to carry out checks without its personnel accessing the underlying material. OpenAI previewed Private Safety Processing in August 2026, noting that existing ZDR-compatible safety checks mostly evaluated individual interactions, which limited the detection of patterns across longer sequences. Under the new architecture, customer content selected by a safety classifier or sampling policy is encrypted and stored in customer-controlled environments such as AWS S3, Azure Blob or Google Cloud Storage. OpenAI retains only an index containing operational metadata and a reference to the stored record, rather than its own plaintext copy. The encrypted data can then be retrieved for automated review inside a hardware-attested safety runtime, which OpenAI says is the only workload capable of decrypting the records and is designed to prevent OpenAI personnel from seeing the underlying customer material.

Private Inference, now in preview, combines confidential computing with verifiable controls. OpenAI has provided few details so far, including which models and API endpoints will support it, what trusted-execution hardware will be used, and how much the service will cost.

The announcement comes amid rising enterprise concerns about data leakage and intellectual-property exposure. Reports in September 2026 said Palantir, Nvidia and Booz Allen Hamilton had restricted or reconsidered their use of advanced models over fears that sensitive information could benefit AI providers. Private Intelligence is intended to draw a harder technical boundary between customers’ sensitive data and OpenAI itself, offering stronger reassurance than contractual promises alone.