Baseten's research arm partners with Hugging Face and Goodfire to build open-model safety standards
Baseten's research arm, Base Labs, announced a partnership with Hugging Face and Goodfire AI to develop safety evaluation and monitoring infrastructure for open-weight models. The initiative aims to address the risk of 'abliteration,' a technique that removes safeguards from open models, with over 6,000 such models currently hosted on Hugging Face. The companies plan to build safety measures directly into model training and deployment, rather than as an afterthought.
Base Labs, created earlier this year, will publish training and monitoring methods for open models. The partnership responds to "abliteration," a technique that strips safety features from open-weight models. Hugging Face currently hosts more than 6,000 such modified models.
Both companies bring substantial funding to the effort. Baseten secured a $1.5 billion Series F in June, reaching a $13 billion valuation. Goodfire, which focuses on making AI decision-making interpretable, raised $150 million in a Series B led by B Capital. The initiative also invites broader developer community participation.
This partnership could establish a baseline for how open-weight AI models are evaluated and deployed, potentially influencing developers and enterprises that rely on these systems. If successful, it may reduce risks associated with modified models while preserving the transparency benefits of open-source AI. However, the framework's effectiveness remains unproven, and adoption across the fragmented open-model ecosystem could be uneven. The outcome may shape broader trust in open AI systems.