OpenAI has released a technical report detailing the performance of gpt-oss-safeguard-120b and gpt-oss-safeguard-20b. These are open-weight reasoning models, post-trained from gpt-oss foundations, designed to interpret and apply a given policy for content labeling. The report provides baseline safety evaluations, comparing these safeguard models against their underlying gpt-oss counterparts.

This development is crucial for the AI industry as it addresses the critical need for robust, scalable safety mechanisms in large language models. By offering specialized models trained specifically for policy adherence, OpenAI empowers developers and researchers to build safer AI applications. The release of these evaluations and models facilitates greater transparency and allows the community to benchmark safety performance, ultimately advancing the standard for responsible AI deployment and mitigating risks associated with model misuse.