Base Labs launches open-weight AI safety standard with Hugging Face and Goodfire

Base Labs, the research division Baseten spun up earlier in 2026, is building safety evaluation and monitoring infrastructure for open-weight AI models alongside Hugging Face and Goodfire AI. The initiative responds directly to the spread of 'abliteration,' a technique that removes safety guardrails from open models — Hugging Face currently lists over 6,000 abliterated models on its platform.

Rather than patching safety on after deployment, the partnership aims to embed it during training. Base Labs framed the effort as a formal 'standard' for transparent, built-in safety controls. 'We believe openness to be an advantage for AI safety,' Baseten posted on X, arguing that open models offer greater visibility into behavior and more actionable safety research than closed-source alternatives.

Goodfire, which specializes in model interpretability — essentially opening the AI 'black box' — is expected to handle the technical 'built-in' safety layer. Baseten, as an AI inference provider, would likely address the deployment side. Technical specifics of how the three parties will collaborate have not been disclosed.

Financially, both lead partners are well-positioned: Baseten raised a $1.5 billion Series F in June 2026, valuing the company at $13 billion, while Goodfire closed a $150 million Series B led by B Capital earlier this year. Base Labs is also issuing an open call for developer contributions to the framework.