When software calls a third-party AI model, the request (the prompt and any data included in it) and the model's response typically pass through the provider's infrastructure. Zero data retention is a commitment from the provider that this data is processed transiently, used only to generate the response, and then not stored or used for any other purpose, including training future models.
This matters because without it, a business sending operational data into an AI feature could unknowingly be contributing that data to a provider's training corpus or leaving it sitting in provider-side logs indefinitely. Zero data retention closes that gap at the provider level.
What to check for
- Whether the commitment applies at the individual account level or only at an enterprise/organisation tier.
- Whether it is a stated data-processing term, not just a general privacy policy.
- Whether it covers both training exclusion and log retention, since a provider can promise one without the other.
Nuromi is direct about the current state of this for its own stack: model calls are made against providers with model training exclusion, and zero-data-retention terms at the organisation level are an active, standing item Nuromi is working through with its model providers rather than something to claim before it is fully in place.