OpenAI is testing a new safety system designed to detect abuse across multiple AI interactions without requiring some customers to give up Zero Data Retention, according to Digital Trends. The company introduced Private Safety Processing in an August 19 announcement, framing it as a way to close a safety gap for privacy-conscious API users whose deployments are built around limiting provider access to prompts and outputs. The problem OpenAI is trying to solve is straightforward but technically awkward. Digital Trends reports that OpenAI says its existing Zero Data Retention safeguards evaluate requests one at a time. That makes it harder to catch abuse patterns that only become visible when multiple interactions are considered together. Private Safety Processing is meant to look across related interactions for suspicious patterns while withholding the underlying customer content from OpenAI employees. When the system detects possible misuse, OpenAI receives what Digital Trends describes as a limited signal about the activity rather than the prompts or responses themselves. For current Zero Data Retention deployments, Digital Trends says the content remains on infrastructure controlled by the customer. OpenAI is also developing a second setup in which encrypted content could reside on OpenAI infrastructure while customers keep control of the encryption keys; under that design, OpenAI would not possess the keys. The program is still early. Digital Trends reports that Private Safety Processing is being tested with early customers, and cites Axios as saying the product is aimed at enterprise and API customers rather than individual ChatGPT subscribers. OpenAI plans to begin rolling it out and publish a technical white paper in September. The announcement puts OpenAI’s approach in contrast with Anthropic’s stated retention model for its Covered Models. According to Digital Trends, Anthropic retains prompts and outputs for 30 days, including for organizations that previously used Zero Data Retention, arguing that the retention window helps safeguards identify attacks that emerge across multiple requests. OpenAI is trying to address the same class of safety risk without requiring eligible customers to accept routine content retention. That distinction matters for companies that adopted Zero Data Retention specifically to reduce how long an AI provider can hold their sensitive data. There are still limits to what can be concluded. Digital Trends notes that OpenAI has not yet publicly demonstrated that Private Safety Processing can provide abuse detection equivalent to systems that retain customer content, and the technical details behind the privacy protections are not yet public. The September white paper is therefore the next important artifact for customers evaluating whether the safety and privacy claims hold up. The system also has a stated exception. Digital Trends reports that OpenAI says images flagged as potential child sexual abuse material can still be retained for legally required manual review and reporting, even under Zero Data Retention. Who benefits: Privacy-sensitive enterprise and API customers stand to benefit if the system works as described. OpenAI also benefits if it can position its safety controls as compatible with Zero Data Retention rather than as a reason to weaken it. Who's exposed: Customers should treat the claims as unproven until OpenAI publishes the technical details and rollout terms. AI providers using retention-based safety approaches may face more pressure to justify why stored prompts and outputs are necessary.