OpenAI has built a safety system called Private Safety Processing that is meant to let enterprise customers use its most advanced AI models without OpenAI storing their data, according to The Decoder. The system is aimed at a hard trade-off for corporate AI deployments: providers want to detect misuse, but customers often want zero data retention for sensitive prompts and outputs. The Decoder reports that Private Safety Processing is designed to detect abuse patterns across multiple related interactions while maintaining zero data retention, or ZDR. In this setup, data is not kept after processing. OpenAI says it receives only a narrow safety signal that includes the type and severity of an activity, rather than the underlying inputs or outputs. The architecture described by The Decoder keeps customer data either on the customer’s own infrastructure or in encrypted form, with the customer retaining the keys. That framing is important: it suggests OpenAI is trying to separate safety monitoring from direct access to customer content, at least as described in the report. Aleah Houze, OpenAI’s head of product policy, told The Decoder that some risks only become visible across multiple conversations. That is the central design tension. A single interaction may not be enough to identify misuse, but storing full conversation content can be unacceptable for customers handling confidential business data. The competitive angle is also explicit. The Decoder reports that Anthropic requires 30 days of data retention for its most powerful models, such as Fable 5. OpenAI’s pitch, as described, is that it can support misuse detection without that retention period, though the available material does not include independent technical validation. OpenAI is expected to release a technical white paper in September, according to The Decoder. Until that paper is public, the implementation details remain limited to the company’s description as reported by the outlet: narrow safety signals, customer-controlled data, and cross-interaction abuse detection without retained customer content. Who benefits: OpenAI benefits if Private Safety Processing makes its advanced models easier to sell into privacy-sensitive enterprise accounts. Customers with strict data-retention requirements may also gain another deployment option, pending technical details. Who's exposed: Rivals that require longer data retention for their most capable models could face pressure in enterprise evaluations, based on The Decoder’s comparison. The risk for OpenAI is that the September white paper will need to substantiate the safety and privacy claims.