OpenAI introduces zero data retention for its frontier models

OpenAI announced that enterprise customers can use its frontier models with zero data retention: requests and responses are not stored on its servers after processing. It addresses the biggest brake on adopting AI for business — the fear that conversations with a model will settle somewhere, end up in training data or leak.

Tellingly, the mode arrived not as an exotic option for a chosen few but as a public offer. The market has matured: large customers spent years putting data handling first on the negotiation agenda, and vendors now compete not only on model quality but on guarantees about information. What used to be dismissed as security-team paranoia has become a line item on the price list.

For a company in Kazakhstan the takeaway is practical: for every AI service your staff uses, establish three things — what gets stored, where it physically resides, and who inside your own company can access what. That last point is usually the most neglected: in 1Cifer, for instance, access is split by position — the accountant and the manager each see their own AI agent and their own data, which removes half the risk before servers even come up.