gpt-oss-safeguard technical report

In three linesOpenAI releases two open-weight reasoning models: gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, post-trained to label content according to provided policies. Technical report covers capabilities and baseline safety evaluations against underlying gpt-oss models.

Summary generated by Claude — human-verified