OpenAI is previewing a service called Private Safety Processing, and it's a direct shot at Anthropic's 30-day data-retention policy. The pitch: eligible enterprises and API customers can use OpenAI's frontier models while maintaining zero data retention — no stored prompts, no stored model responses.
The system can scan multi-turn conversations for potential abuse, but OpenAI employees never see the actual prompts or replies. Customer data can stay on the customer's own servers. Or, if it lives with OpenAI, it's encrypted using keys that only the customer holds — OpenAI doesn't have access.
If the system detects a risk, it returns a limited safety signal to OpenAI, such as which category of suspicious activity was flagged. The raw conversation stays put.