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

Base Labs, the research division of Baseten, announced a new safety infrastructure standard on Wednesday. This initiative involves a partnership with Hugging Face and Goodfire AI, focusing on creating infrastructure for evaluating and monitoring the safety of open-weight models.

The collaboration aims to develop and publish methods for training and overseeing open models, framing their future work as a transparent standard integrated into model development and deployment.

Baseten previously secured a $1.5 billion Series F funding round in June, which elevated its valuation to $13 billion.

Goodfire AI also raised a $150 million Series B round earlier this year, with B Capital leading the investment to advance its model interpretability platform.

Base Labs is the research group established by Baseten, an AI inference provider. Hugging Face is known for hosting open-source AI models.

Goodfire AI specialises in explaining AI model decisions, aiming to demystify the 'black box' of artificial intelligence.

This partnership addresses the ongoing debate surrounding the safety of open-weight models, which can be rendered unsafe by removing safeguards through a technique known as abliteration. Hugging Face currently lists over 6,000 such models, indicating the scale of the challenge.

The companies advocate that openness enhances AI safety by providing greater visibility into model behaviour and offering more effective, transparent controls than closed-source alternatives. The goal is to embed safety directly into the models' design and deployment, rather than adding it on later.

Baseten is extending an open invitation to the broader developer community to contribute to this new framework. The partners aim to cultivate an ecosystem of open models that are both secure and widely accessible.