OpenAI says the data your company sends it won't stick around, even as its models get smarter.
In a Wednesday post titled 'Offering Zero Data Retention for frontier models,' OpenAI told enterprise customers that its zero-data-retention policy will carry over to its next generation of models. The company previewed a new system called Private Safety Processing, designed to scan for misuse across multiple related interactions without OpenAI keeping a permanent copy of customer data. Details on how the detection actually works, and which customers qualify, are still thin.
Enterprise buyers have been wary of sending sensitive data to a model provider that might quietly log it for training or moderation. A misuse-detection system that works without retention is the harder engineering problem: it is easy to promise privacy when you are not looking for abuse, and easy to catch abuse when you are storing everything. OpenAI is betting it can do both at once, which is the real test here.
Whether that holds up under scrutiny from security researchers and privacy regulators is a separate question from whether OpenAI says so in a blog post.