Private Safety Processing (Preview)

A system that lets OpenAI spot misuse across related conversations without staff seeing the content, for enterprise and API customers on zero data retention.

Details

OpenAI's existing safety systems evaluate each interaction on its own; this one looks at patterns across related interactions, which catches someone splitting a request across many conversations to avoid detection
When the automated system flags something, OpenAI receives only a narrow signal naming the category of concern, without the prompts or responses attached
In testing with early enterprise and API customers now, with a broader rollout and a technical paper planned for September
OpenAI paired the preview with a commitment to keep offering zero data retention on frontier models

Best for

Keeping a zero-retention agreement in place on frontier models instead of trading it away for safety monitoring.

When to use it

Legal or security blocked a frontier model because the safety tooling required keeping data.

Who benefits

Enterprise and API buyers in regulated industries who negotiated zero data retention.

Staying informed is great. Staying ahead is better.

Pick a topic and leave with something you created.

Start for Free Bring to Your Team

Get new module drops and weekly AI strategy.

Details

Best for

When to use it

Who benefits

More from OpenAI

Staying informed is great. Staying ahead is better.