OpenAI is previewing a new service called Private Safety Processing for select customers. The automated system watches for potential misuse while retaining none of the customer’s data. It extends the company’s existing Zero Data Retention approach by analyzing inputs and outputs across multiple conversations rather than a single session. An agent performs the monitoring. If triggered, it sends OpenAI a narrowly defined signal about the type of activity. OpenAI can then decide whether enforcement is needed and reach out to the customer, who may choose to share data. The move contrasts with Anthropic’s recently announced policy that retains user data for 30 days on “covered models,” including Mythos-class systems. That policy was intended for safety analysis but has concerned some enterprises handling sensitive information. OpenAI says its personnel never see the underlying content under the new system.
OpenAI is previewing a new service called Private Safety Processing for select customers. The automated system watches for potential misuse while retaining none of the customer’s data. It extends the company’s existing Zero Data Retention approach by analyzing inputs and outputs across multiple conversations rather than a single session. An agent performs the monitoring. If triggered, it sends OpenAI a narrowly defined signal about the type of activity. OpenAI can then decide whether enforcement is needed and reach out to the customer, who may choose to share data. The move contrasts with Anthropic’s recently announced policy that retains user data for 30 days on “covered models,” including Mythos-class systems. That policy was intended for safety analysis but has concerned some enterprises handling sensitive information. OpenAI says its personnel never see the underlying content under the new system.