OpenAI Keeps User Data 30 Days Under New Policy

OpenAI is testing a new privacy tool with a small group of customers, and the timing looks pointed. The company calls it Private Safety Processing: an automated system built to flag potential misuse of its models without storing any of the customer data it scans. OpenAI confirmed the preview this week, framing it as proof that safety monitoring doesn't require holding onto conversations after the fact.

The move lands just months after Anthropic changed its own data policy, a decision that irritated some of its enterprise clients. AI labs are stuck managing a tension that didn't exist a few years ago: models are strong enough now to cause real harm if misused, which pushes companies toward more monitoring, but the customers paying for that access, banks, hospitals, law firms, don't want their sensitive material sitting on a server somewhere waiting to be reviewed.

OpenAI already offers a Zero Data Retention option for select customers, built around agents that operate inside its own infrastructure. Private Safety Processing pushes that idea further, and Anthropic's July policy shift is the backdrop making OpenAI's timing look less like coincidence and more like a direct response.