Baseten, Hugging Face, and Goodfire AI Launch Open-Weight AI Safety Standard

Baseten, through its research division Base Labs, introduced a new safety infrastructure standard for open-weight AI models on September 17, 2026, in partnership with Hugging Face and Goodfire AI. The initiative aims to address safety and transparency in the deployment of open-weight models.

Base Labs will be responsible for developing and releasing procedures for training and monitoring open-weight models. As part of the collaboration, Baseten will incorporate the resulting safety research into its deployment systems, providing live runtime monitoring as a managed service.

Hugging Face, which serves as a platform for hosting open-weight models, is participating in the partnership to facilitate the distribution and hosting of models that comply with the new safety standard. Goodfire AI, a company focused on AI interpretability, is contributing its expertise to help explain model decisions within the safety framework.

According to Baseten, the safety framework is designed as a standard that is integrated directly into the training and deployment processes of open models, rather than being applied after development. The company has also called on the broader developer community to participate in the ongoing development of the framework.

What the numbers show

• Baseten completed a $1.5 billion Series F funding round in June 2026, reaching a $13 billion valuation.
• Goodfire AI raised $150 million in a Series B round earlier in 2026.
• Hugging Face currently lists more than 6,000 abliterated open-weight models.

Baseten’s integration of the safety research outputs will allow users to access live monitoring of deployed open-weight models as a managed service. This approach is intended to make safety measures a foundational part of model deployment for organizations using Baseten’s infrastructure.

Hugging Face’s involvement focuses on ensuring that open models distributed through its platform adhere to the new safety standard. The company’s platform currently hosts thousands of open-weight models, including those with removed safeguards.

Goodfire AI’s participation centers on providing tools and methods for interpreting and explaining the decisions made by AI models. The company’s recent funding round supports its ongoing work in model interpretability, which is a key component of the safety initiative.

Baseten’s invitation to the developer community is intended to encourage wider participation in shaping the safety framework. The company described the standard as transparent and embedded within the model lifecycle, aiming for broader adoption across the open-weight AI ecosystem.

* This article is based on publicly available information at the time of writing.