OpenAI has announced a new Private Safety Processing system designed to maintain zero data retention for enterprise customers while identifying complex security threats across multiple user interactions.

Key Points

• OpenAI’s new system allows enterprise and API customers to maintain zero data retention while enabling the company to detect coordinated cyber threats across multiple prompts.
• The technology uses automated, encrypted review processes that prevent OpenAI staff from accessing or viewing sensitive customer content during safety monitoring.
• Customers can choose to keep data on their own infrastructure or use OpenAI-hosted storage secured by customer-held encryption keys.
• Early testing partners for the initiative include Microsoft, Databricks, Glean, and Abridge, with a full rollout and technical white paper expected in September.
• The policy excludes consumer ChatGPT plans and includes a mandatory exception for reporting child sexual abuse material as required by United States law.

Why it Matters

This development highlights a growing industry divide regarding how AI companies balance user privacy with the need to monitor for sophisticated security risks. By offering an alternative to the mandatory data retention policies recently adopted by competitors like Anthropic, OpenAI aims to secure its position as the preferred provider for highly regulated sectors.