Technical Report: Performance and baseline evaluations of gpt-oss-safeguard-120b and gpt-oss-safeguard-20b

gpt-oss-safeguard-120b and gpt-oss-safeguard-20b are two open-weight reasoning fashions post-trained from the gpt-oss fashions and educated to purpose from a supplied coverage in an effort to label content material beneath that coverage. In this report, we describe gpt-oss-safeguard’s capabilities and present our baseline security evaluations on the gpt-oss-safeguard fashions, utilizing the underlying gpt-oss fashions as a baseline. For extra details about the event and structure of the underlying gpt-oss fashions, see the unique gpt-oss mannequin mannequin card⁠.