Base Labs Launches Open-Weight AI Safety Partnership with Hugging Face and Goodfire
Summary
- • Baseten's Base Labs creates safety standard for open-weight AI models with major partners
- • Partnership targets abliteration threat: 6,000+ safety-bypassed models on HuggingFace
- • Framework built into model training from the start, not bolted on after deployment
- • Baseten ($13B valuation) and Goodfire ($150M raised) lead well-funded initiative
Details
Base Labs, HuggingFace, and Goodfire form open-weight AI safety standard
Baseten's research arm Base Labs launched a new safety infrastructure initiative partnering with Hugging Face and Goodfire AI to build evaluation and monitoring for open-weight models.
Over 6,000 abliterated models with removed safety features listed on HuggingFace
Abliteration is a rising technique that strips safety guardrails from open models. HuggingFace currently lists over 6,000 such modified models, illustrating the massive scale of the open-source safety threat.
Safety standard built into model training process, not added as afterthought
Base Labs frames the standard as transparent and embedded in how models are trained and deployed, contrasting with current approaches that layer safety features on top of existing models.
Baseten raised $1.5B Series F in June 2026 at $13B valuation
The June fundraise vaulted Baseten into the top tier of AI infrastructure companies and funds the Base Labs safety initiative.
Goodfire raised $150M Series B led by B Capital for model interpretability
Goodfire specializes in opening AI's 'black box' to explain how models make decisions, making it the key technical contributor for the interpretability layer of the partnership.
Open call to developer ecosystem to contribute to the safety framework
Baseten is inviting broader developer participation, aiming to build 'an ecosystem of open models that are safe and accessible to all.'
Details on the Base Labs open-weight AI safety partnership announced September 17, 2026.
What This Means
Baseten's Base Labs is creating a new industry standard for open-weight model safety, directly addressing a growing crisis where thousands of models have had their safety features stripped through 'abliteration.' By partnering with Hugging Face and Goodfire and inviting broader developer participation, this initiative could define how open-source AI safety is built and verified. The approach — embedding safety into training rather than applying it as an afterthought — represents a meaningful architectural shift. With Baseten valued at $13B and Goodfire well-capitalized, this is not a paper partnership but a funded effort with real resources behind it.