Base Labs, the AI research division of inference provider Baseten, has teamed up with Hugging Face and Goodfire AI to create a new safety evaluation and monitoring standard tailored to open-weight models. The collaboration aims to embed safety controls into open-source AI from the ground up, rather than patching protections afterward.
Why This Partnership Matters
Open-weight models—those whose full parameter sets are accessible—offer transparency and adaptability, but they also carry risks. A growing concern is a technique now frequently used called “abliteration,” which involves stripping safety layers away, leaving models more vulnerable to misuse. Hugging Face currently hosts thousands of models that have been abliterated. The risk amplifies when models are modified or applied in ways not originally intended.
Base Labs plans to develop and publish methods for both training and ongoing monitoring of such models. The goal is to define a standard for open AI models that’s transparent, integrated into development pipelines, and consistent. This is aimed at bridging a gap where many safety tools have traditionally been added only after models are fully built.
Strengths of the Partners & Funding Behind Them
Baseten itself is a well-funded inference platform; in June it secured a $1.5 billion Series F, valuing the company at $13 billion. Goodfire AI, which focuses on interpretability—helping reveal how models make decisions—raised $150 million in a Series B earlier this year, signaling broader market investment in tools that open up AI’s “black box” behavior. Hugging Face brings deep experience hosting and providing infrastructure for open-source models.
Operationally, these companies haven’t released technical blueprints yet. Goodfire emphasized that safety should not be an afterthought—it should be built in by those serving open-weight models. Meanwhile, Base Labs is calling on the wider developer community to contribute to this emerging framework, pushing for an ecosystem where open models are both safer and more accessible.
This announcement arrives at a moment when open solutions are under scrutiny. The question of whether openness heightens risk—or provides more room for responsible oversight—is central to debates in AI governance. Base Labs and its partners are betting on the latter, arguing that open models can benefit from stronger external audit, more transparent behavior, and actionable controls.
What this means going forward: If this initiative takes hold, it could shift how open-weight models are built, deployed, and regulated. Developers may soon need to adhere to safety standards from the start, rather than retrofitting protections. Regulatory bodies and AI ethics stakeholders will likely watch closely, as will those who view open models as key to AI’s future innovation.
