It's not as flashy as a big model release, but enterprise controls are hugely important – and play a massive role in how AI gets adopted. I'm glad we spend a lot of calories thinking it through.
ZDR, abuse monitoring, and human review have come up _countless_ times with enterprises. It's often the only blocker for them to roll out AI across their firm.
With Private Safety Processing, the outputs of safety processing is stored in a way such that no OpenAI personnel gets access to retained customer info. The customer stays in control of the data.
If this is something interesting to you, reach out and let us know! We're rolling this out very soon.
We will continue to offer Zero Data Retention for frontier models. As AI takes on longer, more autonomous work and delivers greater value to businesses, safety ...