OpenAI wants to offer enterprise customers its most advanced AI models without storing their data, while still catching misuse. The company built a safety system called "Private Safety Processing" to make that work.
The system detects abuse patterns across multiple related interactions while maintaining zero data retention (ZDR), meaning no data is kept after processing. OpenAI says it only receives a narrow safety signal that includes the type and severity of an activity, without seeing the actual inputs or outputs.
Customer data remains on the customer’s own infrastructure or is stored in encrypted form, with the customer retaining the keys. Aleah Houze, Head of Product Policy at OpenAI, said that risks often only become apparent over the course of multiple conversations. Competitor Anthropic, on the other hand, requires 30 days of data retention for its most powerful models, such as Fable 5. According to OpenAI, a technical white paper is expected to be released in September.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
Subscribe now