What is OpenAI Private Intelligence, and do you have to switch it on?

Every business nervous about AI has the same worry. My data goes in, and then what? Private Intelligence is OpenAI's pitch to those people, previewed at DevDay.

It has two ingredients. First, zero data retention, meaning the provider does not keep your prompts and outputs. Second, automated safety checks that run without OpenAI staff seeing the content. OpenAI also says business data is not used for training unless the customer opts in.

Is it something you flip on in every chat? Going by how it is described, no. It looks like an account level setup, the kind of thing an admin configures once, not a button you press per conversation.

Hmm, safety checks on content that nobody looks at. How is that even supposed to work?

By machine. The checks are automated, so the whole idea is that software scans for problems and people do not read your data. How much comfort that gives a heavily regulated business is the real test.

There is a second piece coming this autumn in preview: Private Inference. The model would run inside confidential computing hardware, a sealed enclave that not even the cloud operator can open, with controls you can verify. For hospitals, banks and law firms, that could decide whether AI gets to touch client data at all.