Base Labs launches an open-weight AI safety partnership with Hugging Face and Goodfire

Baseten said on Wednesday that it and its research arm, Base Labs, have launched a new safety infrastructure standard for open-weight AI models, working with Hugging Face and Goodfire AI to develop evaluation and monitoring capabilities. The stated aim is to create tools and methods that are integrated into model training and deployment rather than added afterward. Baseten described openness as beneficial to safety because it increases visibility into model behavior and enables transparent controls. The announcement comes amid growing concern about the risks associated with open-weight models, particularly a technique called abliteration that can remove built-in safeguards. Hugging Face, which hosts open-source models, currently lists more than 6,000 abliterated models, underscoring the scale of the issue the partners are addressing. Specific technical details of the partnership have not been disclosed. Goodfire, which focuses on model interpretability and explaining model decisions, responded to Baseten’s post by emphasizing that safety should be embedded in open models and provided by those who serve them, suggesting its role may center on transparency and explainability within the proposed standard. Baseten has recently expanded its market position after a $1.5 billion Series F in June that lifted its valuation to $13 billion. Goodfire AI has also secured significant backing, raising $150 million in a Series B round led by B Capital. Baseten said it will open the effort to contributions from the broader developer ecosystem as it works to establish the standard and related tooling.