PANews reported on October 29th that OpenAI today released the open-source security inference model gpt-oss-safeguard ( 120b , 20b ), allowing developers to provide custom policies for content classification during inference, with the model outputting conclusions and inference chains. This model is fine-tuned based on the open-weight gpt-oss , licensed under the Apache 2.0 license, and can be downloaded from Hugging Face . Internal evaluations show that it outperforms gpt-5-thinking and gpt-oss in multi-policy accuracy, and its performance on external datasets is close to Safety Reasoner . Limitations include: traditional classifiers still outperform in scenarios with a large number of high-quality annotations, and higher inference time and computational cost. ROOST will establish a model community and release technical reports.