What changed

OpenAI released a Safety Usage Dashboard that groups blocked Responses API requests by the safety_identifier supplied for an end user. The view makes provider-side safety blocks visible at organization level instead of leaving teams to infer them from individual request failures.

Why it matters

API teams can separate policy-block patterns from ordinary application errors and investigate whether one integration, user segment or identifier is producing repeated denials.

What to do

Confirm that every production Responses request carries a stable, privacy-reviewed safety_identifier, then compare dashboard blocks with application logs and appeal or support records before changing policy or user access.

Caveat

The dashboard reports OpenAI's blocking decisions, not ground truth about user intent or harmfulness. Access is permission-gated, and identifiers still require privacy, retention and false-positive controls.