Zero data retention is a service mode in which prompts and generated outputs are not stored after inference. In practice, the provider processes the request, returns the result, and discards the content instead of keeping it for later reuse, training, or customer review, depending on the platform’s exact setup.
In cyber security, this matters because prompts often contain source code, incident details, credentials, or other sensitive business data. Zero data retention can reduce the amount of data exposed if a provider environment is breached or if retained content is subpoenaed or mishandled. It is not a complete privacy guarantee, though: organizations still need to check whether logs, telemetry, billing records, error traces, or customer-managed copies are retained elsewhere. Defenders often pair this mode with strict access controls, encryption, and limited tool permissions when using AI for code review, analysis, or incident response.