Baseten’s Base Labs Targets 6,000 Abliterated AI Models With Open Safety Standard
Baseten, Base Labs, Hugging Face and Goodfire build open-weight AI safety standards as over 6,000 abliterated models expose rising risks across open source AI.
Summary
As of September 17, 2026, Baseten and its research arm, Base Labs, have launched an open-weight AI safety initiative with Hugging Face and Goodfire AI. Base Labs, created earlier in 2026, will publish transparent training and monitoring methods intended as a standard embedded in model training and deployment, rather than added afterward. Baseten argues openness exposes model behavior and converts safety research into actionable controls.
The focus is abliteration, a growing technique for stripping safeguards from open-weight models; Hugging Face hosts more than 6,000 abliterated models. Technical roles remain undisclosed. Goodfire says companies serving open models should build safety into them, while its interpretability platform explains model decisions. AI inference provider Baseten raised a $1.5 billion Series F in June 2026, reaching a $13 billion valuation. Goodfire raised a $150 million Series B led by B Capital earlier in 2026. Baseten is inviting developers to contribute to an ecosystem of safe, accessible open models.
Positives
- Base Labs will publish transparent methods for training and monitoring open models.
- Hugging Face and Goodfire AI are joining Baseten to develop safety evaluation and monitoring infrastructure.
- Baseten raised a $1.5 billion Series F in June 2026, reaching a $13 billion valuation.
- Goodfire raised a $150 million Series B led by B Capital earlier in 2026.
- Baseten has invited the wider developer ecosystem to contribute to the framework.
Risks & concerns
- Hugging Face already lists more than 6,000 abliterated models with safeguards removed.
- Abliteration allows open-weight models to be made dangerous by stripping their built-in protections.
- Baseten, Hugging Face and Goodfire have not disclosed how the partnership will work technically.
- The proposed safety standard is still being developed and requires contributions from the broader developer ecosystem.