OpenAI Unveils Private Safety Processing for Zero Data Retention AI Deployments
Brief
OpenAI has previewed Private Safety Processing (PSP), a safety-monitoring architecture designed to identify misuse patterns across related AI interactions while preserving its Zero Data Retention (ZDR) commitment for eligible API customers.
The approach addresses a growing enterprise dilemma: stronger frontier-model safeguards often require broader behavioral context, but retaining sensitive prompts and outputs can conflict with privacy, security, and regulatory obligations.
Under OpenAI’s ZDR model, prompts and model responses are not retained after processing, customer content is unavailable for review by OpenAI personnel, and enterprise data is not used for model training unless the customer explicitly opts in.