Baseten and partners launch safety standard for open-weight AI models

Baseten, an AI inference platform, unveiled a collaborative safety framework for open-weight models this week through its Base Labs research division, partnering with Hugging Face and Goodfire AI to create evaluation and monitoring systems designed to protect openly available models from misuse.

The timing reflects growing concerns about the vulnerability of open-source models to manipulation. A technique called abliteration — which strips away safety mechanisms built into models — has proliferated rapidly. The problem is significant enough that Hugging Face currently hosts more than 6,000 models that have had their safeguards removed in this manner.

Building safety into architecture, not tacked on#

The initiative centers on a conceptual shift: rather than applying safety measures as external patches after models are created, Baseten's approach integrates protective mechanisms during model development and deployment. The company positions this as a transparent "standard" for the open-source AI ecosystem, contrasting it with retrofitted solutions that appear grafted onto existing systems.

Baseten framed the philosophy in a public statement: "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 idea is that transparency itself becomes a safety tool, allowing researchers and developers to understand how models function and intervene more effectively.

Partnership roles and contributions#

While the exact technical workings remain undisclosed, each partner brings distinct expertise. Goodfire specializes in model interpretability — making AI decision-making processes more understandable — positioning it as the likely contributor for embedding safety mechanisms into model architecture. Hugging Face provides the infrastructure and community access through its massive open-source model repository.

Goodfire reiterated the core principle in its response to the announcement: safety must be integral to open models and managed by those who deploy them. This philosophy echoes broader industry sentiment that security cannot be an afterthought in democratized AI systems.

Market momentum#

Both Baseten and Goodfire are well-funded enterprises. Baseten closed a Series F round in June valued at $1.5 billion, bringing the company's valuation to $13 billion. Goodfire secured $150 million in Series B funding earlier in 2026, backed by B Capital, to accelerate its interpretability platform.