gpt-oss-safeguard

gpt-oss-safeguard is OpenAI's open weight reasoning model tuned for policy classification. You pass your written policy and the content together, and instead of returning a bare label it reasons through the policy and explains why the content does or doesn't violate it. That makes it slower and more expensive than a small classifier but far better on nuanced or novel policies, and the rationale is useful for reviewer training and appeals.

It comes in two sizes and runs anywhere gpt-oss runs. OpenAI positions it as the tool you reach for when a fixed category classifier isn't enough and you're willing to spend compute on judgment.