Three companies join forces on AI safety

Baseten's Base Labs research arm partnered with Hugging Face and Goodfire AI to build safety evaluation and monitoring infrastructure for open-weight models.

The abliteration problem is massive

Over 6,000 abliterated models are hosted on Hugging Face. Abliteration removes safety safeguards from open-weight models, making them dangerous.

Safety built in, not bolted on

Base Labs will develop methods for training and monitoring open models as a transparent standard built into deployment, rather than added afterward.

“We believe openness to be an advantage for AI safety.”

Openness enables better safety controls

Open models provide more visibility into behavior and allow safety research to become actionable and transparent controls, unlike closed-source alternatives.

“Openness provides more visibility into the behavior of models and greater means of turning safety research into actionable controls.”

Goodfire brings interpretability expertise

Goodfire AI, which specializes in explaining how models make decisions, will likely handle building safety directly into open models as part of the partnership.

“Safety must be built into open models and provided by those who serve them.”

Well-funded partners drive the effort

Baseten raised 1.5 billion dollars in June at a 13 billion dollar valuation, while Goodfire raised 150 million dollars in Series B funding earlier this year.