OpenAI unveils Private Safety Processing with zero data retention

OpenAI built Private Safety Processing to detect abuse across interactions while maintaining zero data retention, receiving only a narrow safety signal. Customer data stays on their infrastructure or encrypted with customer keys. Anthropic requires 30 days retention for models like Fable 5. A technical white paper arrives in September.