Baseten announced on Wednesday that it has launched a new safety infrastructure standard together with its research arm Base Labs, in partnership with Hugging Face and Goodfire AI. The initiative aims to build evaluation and monitoring infrastructure for open-weight models.
Why this matters now
The announcement comes amid growing concern over the safety of open-weight models. A key risk is that protections baked into models can be removed using techniques referred to in the announcement as "abliteration." The scale of the issue is large: Hugging Face, which hosts open-source AI models, currently lists more than 6,000 abliterated models.
What the partners will do
Base Labs — the research group Baseten established earlier this year — will develop and publish methods for training and monitoring open models. Baseten frames the work as creating a "standard" for open models that is transparent and integrated into how models are trained and deployed, rather than added on afterwards.
In a post on X, Baseten stated: “We believe openness to be an advantage for AI safety. Openness provides more visibility into the behavior of models and, most importantly, greater means of turning safety research into actionable and transparent controls than closed-source.”
Technical role and Goodfire AI
The companies have not revealed technical implementation details. In a reply to Baseten’s announcement, Goodfire AI said: “Safety must be built into open models and provided by those who serve them.” Goodfire specializes in explaining model behavior and opening AI’s "black box," making it the most likely partner to lead the "built-in" safety work.
Financial context
Baseten, an AI inference provider, raised $1.5 billion in a Series F round in June, bringing its valuation to $13 billion. Partner Goodfire AI also raised significant capital this year, securing a $150 million Series B led by B Capital to advance its model interpretability platform.
Invitation to the developer community
Baseten is publicly inviting the wider developer ecosystem to contribute to the framework. The company said the goal is to build an ecosystem of open models that are safe and accessible to all.
Summary
The joint effort seeks to address risks from abliterated open models by embedding safety evaluation and monitoring into model training and deployment. While technical specifics remain to be disclosed, Baseten, Base Labs, Hugging Face and Goodfire AI aim to create a transparent, community-driven standard and are calling for broader participation.
