Baseten announced new safety infrastructure standards on Wednesday alongside its Base Labs research division and partnered with Hugging Face and Goodfire AI to build a safety assessment and monitoring infrastructure for open weight models.
The announcement comes amid debate over the safety of open weight models. A technology known as abliteration is emerging, and removing safeguards can be dangerous. The scale of the problem is enormous. Hugging Face, which hosts open source AI models, currently has over 6,000 defunct models listed.
Base Labs is a research group that Baseten launched earlier this year to develop and publish methods for training and monitoring open models. The company is framing its future efforts as an open model “standard.” This is transparent and built into the way models are trained and deployed, rather than being added as an afterthought.
“We believe openness is an advantage for the safety of AI,” the company said at X. “Openness provides greater visibility into model behavior and, most importantly, a better means of giving safety research more practical and transparent control than closed source.”
The companies have not disclosed how the partnership will work technically, but Goodfire said in a reply to Baseten’s post that the goal is that “safety must be built into the open model and provided by the people who serve it.” Goodfire specializes in opening up the “black box” of AI to explain how models make decisions, making it a prime candidate for the “embedded” part.
Baseten, an AI inference provider, raised $1.5 billion in Series F in June, giving it a $13 billion valuation. Partner Goodfire AI is similarly well-capitalized, having raised $150 million in a Series B led by B Capital earlier this year to advance its model interpretability platform.
Looking to the future, Baseten is inviting the broader developer ecosystem to contribute to the framework. “Together, we are building an open model ecosystem that is safe and accessible to everyone,” the company said.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
