OpenAI has announced the launch of a completely new open-source security model suite gpt-oss-safeguard, designed to provide AI systems with more flexible, transparent, and auditable security classification capabilities. The model includes two versions, 120 and 20, and is released under the Apache 2.0 license, allowing developers to freely use, modify, and integrate it.

Unlike traditional security classifiers, gpt-oss-safeguard supports "real-time policy interpretation", which means that when security or content rules change, the model can adapt and update instantly without retraining. This mechanism significantly reduces the maintenance costs of security systems, enabling enterprises and organizations to respond faster to evolving compliance and content security needs.

OpenAI, ChatGPT, artificial intelligence, AI

In terms of transparency, OpenAI states that the architecture of gpt-oss-safeguard allows developers to directly view the model's decision-making process, making it easier to understand its judgment logic and facilitating auditing and optimization. This design addresses long-standing concerns about AI's "black box" problem and provides a new technological paradigm for building a trustworthy AI security ecosystem.

Notably, gpt-oss-safeguard is built upon OpenAI's own open-source model gpt-oss and is launched as a collaborative effort between OpenAI and the ROOST platform (an open-source community focused on AI security, safety, and governance infrastructure). OpenAI says that the goal of this project is to promote a more open and responsible AI security standardization process globally.