OpenAI has released an image classifier intended to help approved organisations review suspected child sexual abuse material, including AI generated images. The company says the tool is available through its API at no cost, but access requires an application, organisational verification, programme onboarding and acceptance of specific terms.
The classifier returns scores that can help trained reviewers prioritise material. OpenAI explicitly says it does not replace human judgment: participating organisations remain responsible for reviewing the result, deciding what action to take and meeting any reporting obligations. That boundary is crucial in a high-risk context where both false negatives and false positives can cause serious harm.
The launch creates a wider governance question for safety tooling. A score is a triage signal, not a legal conclusion or an instruction to report. Deployers need evidence on calibration, documented thresholds, reviewer training, access controls, audit trails and secure handling of extremely sensitive data. The company has not published a general benchmark in the release note, so effectiveness should not be inferred from availability alone.
