Baseten has introduced a new safety infrastructure standard through its research division, Base Labs. The initiative involves a partnership with Hugging Face and Goodfire AI to create monitoring and evaluation tools specifically for open-weight artificial intelligence models.
The collaboration addresses the risks associated with abliteration, a technique used to remove safety safeguards from models. Hugging Face currently hosts more than 6,000 models that have undergone this process. Base Labs intends to develop and publish methods for training and monitoring that integrate safety directly into the deployment process.
Goodfire AI will contribute its expertise in model interpretability to help explain how AI systems make decisions. The partners aim to establish a transparent framework for safety that is built into the architecture of open models rather than added as an external layer.
Baseten recently secured $1,5 billion in a Series F funding round, which brought its total valuation to $13 billion. Its partner, Goodfire AI, raised $150 million in a Series B round led by B Capital earlier this year to support its interpretability platform.
Base Labs has issued an open invitation to the developer community to contribute to the development of this safety framework. The goal is to foster an ecosystem where open-source models remain both accessible and secure for all users.
