Base Labs, the research division of Baseten, has introduced a new initiative to strengthen safety practices for open-weight AI models. The program brings together Hugging Face and Goodfire to develop shared infrastructure for evaluating, monitoring and improving openly available models.
Safety Designed Into the Model Lifecycle
The partnership aims to make safety a core component of how open models are trained, deployed and maintained. Rather than applying controls after release, Base Labs plans to publish transparent methods that developers can integrate throughout the model lifecycle.
Open-weight models offer researchers and developers greater visibility into how AI systems function. However, the growing practice of removing built-in protections, often described as abliteration, has highlighted the need for clearer technical standards. Hugging Face currently lists more than 6,000 models associated with this approach.
Goodfire is expected to contribute its work in model interpretability, helping developers better understand how AI systems produce outputs and where safety controls can be strengthened. Baseten, meanwhile, will support the operational side through its AI inference infrastructure and Base Labs research efforts.
An Open Call for Developers
The companies are inviting the wider developer community to help shape the framework. Their goal is to create practical, accessible tools that support responsible experimentation while preserving the collaborative advantages of open AI development.
By pairing transparency with built-in evaluation and monitoring, this initiative could help establish a more reliable foundation for the next generation of open AI systems.
