What changed
OpenAI released 20B and 120B gpt-oss-safeguard checkpoints under Apache 2.0, designed to classify content by reasoning over developer-provided policies at inference time.
Why it matters
Teams can inspect, customize and self-host a safety layer without retraining a classifier whenever policy language changes.
What to do
Evaluate it on your actual policy and edge cases, compare against a dedicated classifier, and route only uncertain items to the slower reasoner.
Caveat
OpenAI says dedicated classifiers can outperform it with enough labeled data, while reasoning adds compute, latency and potentially reviewable sensitive traces.