Base labs launches open-weight safety partnership

Baseten has launched a new safety infrastructure standard alongside its Base Labs research arm. The company is partnering with Hugging Face and Goodfire AI to build safety evaluation and monitoring infrastructure for open-weight models, which are AI models whose underlying code and weights are publicly accessible.

The challenge of modification

The announcement arrives during an ongoing industry debate regarding the safety of open models. These systems can sometimes have their safety guardrails removed through a rising technique known as abliteration. The scale of this issue is significant, with Hugging Face currently hosting over 6,000 abliterated models on its platform.

Base Labs, a research group established by Baseten earlier in the year, plans to develop and publish new methods for training and monitoring open systems. Rather than treating safety as an afterthought, the company aims to establish a transparent standard built directly into how models are trained and deployed.

Collaborative safety

While technical details of the partnership remain unannounced, Goodfire AI is expected to contribute its expertise in model interpretability. Goodfire specializes in opening AI's internal mechanics to explain how systems make decisions.

Baseten operates as an AI inference provider, delivering a platform for running machine learning models in production. The company recently raised a $1.5 billion Series F funding round in June, bringing its valuation to $13 billion. Goodfire AI is also backed by substantial capital, having secured a $150 million Series B funding round earlier in the year.

Looking ahead, the organizers have issued an open call inviting the broader developer ecosystem to contribute to the new framework.