Base Labs has launched a safety infrastructure standard for open-weight AI models in collaboration with Hugging Face and Goodfire, addressing concerns about the rising technique of abliteration which can compromise model safety.

The partnership aims to integrate safety evaluation and monitoring directly into the training and deployment of open models, ensuring that transparency and safety research translate into actionable controls.

With over 6,000 abliterated models on Hugging Face alone, the scale of the challenge is daunting. Baseten sees openness as a potential advantage for AI safety, providing greater visibility into model behavior and enabling more effective safety measures.

Goodfire, which specializes in making AI models more interpretable, is likely to play a key role in embedding these safety standards. Baseten’s recent $1.5 billion funding round suggests they are well-positioned to lead this initiative.

The broader developer ecosystem is invited to contribute to the framework, aiming to build an open, safe, and accessible model ecosystem for all.