Baseten unveiled a new safety infrastructure standard on Wednesday, introduced in conjunction with its Base Labs research division. The initiative involves collaborations with Hugging Face and Goodfire AI, aimed at creating safety evaluation and monitoring systems for open-weight models.

This announcement arrives during ongoing discussions about the safety of open-weight models, which can be rendered hazardous by stripping away their protective measures via a growing method called abliteration. The problem's magnitude is considerable: Hugging Face, a platform for hosting open-source AI models, presently features more than 6,000 abliterated models.

Base Labs, the research unit Baseten established earlier this year, will create and release approaches for training and overseeing open models. The organization describes its forthcoming efforts as a "standard" for open models that is clear and integrated into the training and deployment processes, instead of being added as an afterthought.

"We believe openness to be an advantage for AI safety," the company stated on X. "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."

The specifics of how the partnership will function technically remain undisclosed, but Goodfire outlined the objective in a response to Baseten's post: "Safety must be built into open models and provided by those who serve them." Goodfire, which focuses on revealing AI's "black box" to clarify how models arrive at decisions, seems the most probable fit for the "built into" component.

Baseten, a provider of AI inference services, secured a $1.5 billion Series F in June, elevating its valuation to $13 billion. Its partner, Goodfire AI, is also well-funded, having obtained a $150 million Series B headed by B Capital earlier this year to further its model interpretability platform.

Moving forward, Baseten is issuing a public invitation to the wider developer community to add to the framework. "Together, we are building an ecosystem of open models that are safe and accessible to all," the company remarked.