Hugging Face lists more than 6,000 abliterated models. That number is the backdrop for a new open-weight safety effort announced on Wednesday, Sept. 16, by Baseten, an AI inference provider, together with Hugging Face and Goodfire AI.
Abliteration is a technique for removing the safeguards from an open-weight model, which can leave it dangerous. In plain terms, anyone who downloads the weights of an open model can strip out the guardrails its makers put in. Baseten says the answer is to build safety into open models from the start rather than add it afterward.
What Base Labs will do
The work sits with Base Labs, a research group Baseten created earlier this year. It will develop and publish methods for training and monitoring open models, and Baseten calls the result a “standard”: transparent, and built into how models are trained and deployed. Think of seatbelts. You can fit them at the factory or hope someone bolts them in later.
“We believe openness to be an advantage for AI safety,” Baseten said 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.”
How the partnership will work technically has not been disclosed. Goodfire, which specializes in opening AI’s “black box” to explain how models make decisions, replied to Baseten’s post with its own view of the goal: “Safety must be built into open models and provided by those who serve them.” TechCrunch noted that Goodfire is the likeliest candidate for the “built into” part.
Money and an open call
Neither partner is short of funding. Baseten raised a $1.5 billion Series F in June, which put its valuation at $13 billion. Goodfire raised a $150 million Series B led by B Capital earlier this year to advance its model interpretability platform.
Baseten has also put out an open call for developers to contribute to the framework. “Together, we are building an ecosystem of open models that are safe and accessible to all,” the company said.
