OpenAI has launched a preview version of Private Safety Processing for select customers, an upgraded safety solution built on its existing zero data retention policy. It uses automated systems to monitor potential misuse of AI models across multiple conversations without storing customer data, enabling detection of malicious activities such as cyberattacks and code generation even when requests are split across sessions. The system generates alerts automatically, allowing OpenAI to assess risks and contact customers if further action is needed (with customers able to decide whether to share additional data). This service addresses the challenge of balancing privacy protections with security requirements for enterprises that refuse to share sensitive information. It also responds to Anthropic’s recent policy allowing 30 days of user data retention for certain models. As AI companies compete in capabilities, security, and commercialization, privacy and security governance have become key differentiators, with both firms preparing for IPOs.