Base Labs, Hugging Face, and Goodfire launch open-weight AI safety partnership

Baseten’s Base Labs, a research group launched earlier this year, announced a partnership with Hugging Face and Goodfire to co-develop and publish methods for training and monitoring open-weight AI models. The collaboration frames its work as a safety standard that is integrated into model development and deployment processes, rather than treated as an afterthought.

The announcement comes as open-weight models face scrutiny due to a technique called abliteration, which can remove safeguards from models. Hugging Face, a platform hosting open-source AI models, currently lists over 6,000 abliterated models, underscoring the scale of the safety challenge.

Base Labs and its partners argue that openness can enhance AI safety by providing greater visibility into model behavior and enabling more actionable safety research. They assert that transparent controls are more feasible in open ecosystems compared to closed-source environments.

The partnership has not disclosed technical specifics about how the safety methods will be implemented. Goodfire, which focuses on model interpretability, is positioned to contribute to the "built-in" safety components of the framework.

Baseten, an AI inference provider, raised a $1.5 billion Series F in June, while Goodfire raised a $150 million Series B earlier this year, indicating substantial financial backing for the initiative. The companies are also inviting contributions from the broader developer ecosystem to shape the framework.

Baseten’s Base Labs was spun up by Baseten, a company that provides AI inference services. The partnership’s stated goal is to build an ecosystem of open models that are both safe and accessible.