Baseten, Hugging Face and Goodfire AI launch a partnership on the safety of open AI models

Baseten has launched a partnership through its research division Base Labs with Hugging Face and Goodfire AI, focused on building safety evaluation and monitoring infrastructure for open (open-weight) models. According to the companies, the goal is to create a transparent safety standard that will be built into models during training and deployment, rather than added afterward.

The initiative responds to the discussion about the safety of open models, which can be stripped of safety measures using a technique called abliteration. According to the article, Hugging Face, which hosts open models, had recorded more than 6000 such abliterated models on its platform. Base Labs, a research group founded by Baseten earlier in 2026, is to develop and publish methods for training and monitoring open models. Baseten stated on X that it views openness as an advantage for AI safety because it provides greater insight into model behavior and better tools for translating safety research into concrete and transparent controls than closed models do. Goodfire AI, which specializes in model interpretability, stated in response that safety must be built into open models by those who operate them.

Technical details of how the partnership will work have not been disclosed. Baseten, an AI inference provider, raised 1.5 billion dollars in Series F funding in June this year, bringing its valuation to 13 billion dollars. Goodfire AI raised 150 million dollars in Series B funding led by B Capital earlier in 2026. Baseten says it is inviting the broader developer community to contribute to the future framework.