Base Labs forms open-model safety partnership with Hugging Face and Goodfire
Baseten has launched a safety infrastructure initiative through its research group, Base Labs, in partnership with Hugging Face and Goodfire AI. The companies say they will develop evaluation and monitoring infrastructure for open-weight models, while Base Labs will create and publish methods for training and overseeing them.
The initiative comes amid concerns that safeguards can be removed from open-weight models through a technique known as abliteration. Hugging Face currently lists more than 6,000 abliterated models. Baseten said the planned framework is intended to make safety transparent and part of model training and deployment, rather than an addition made later. The company argues that openness can improve safety by making model behavior more visible and enabling research to become transparent controls.
The partners have not disclosed the technical structure of their collaboration. Goodfire focuses on interpreting model behavior, while Baseten provides AI inference services. Baseten is inviting developers to contribute to the framework and says the broader goal is an ecosystem of open models that are both safe and accessible. Baseten raised a $1.5 billion Series F in June, reaching a $13 billion valuation; Goodfire AI raised a $150 million Series B earlier this year.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.