Base Labs, the safety research arm spun out of Baseten, partnered with Hugging Face and Goodfire on September 17, 2026 to build safety evaluation and monitoring infrastructure for open-weight models.
- This guide provides comprehensive, actionable information
- Consider your specific workflow needs when evaluating options
- Explore our curated LLMs tools for specific recommendations
What actually shipped
On September 17, 2026, Base Labs — the AI safety research arm spun out of Baseten earlier this year — announced a partnership with Hugging Face and Goodfire to build safety evaluation and monitoring infrastructure for open-weight models, reported first by TechCrunch. Goodfire is contributing its model-interpretability work; the stated goal is to develop and publish methods for training and monitoring open models so safety is built in during development rather than patched on afterward. Base Labs put it directly: "We believe openness to be an advantage for AI safety. Openness provides more visibility into the behavior of models."
The number that explains why this exists
Hugging Face's own count, cited in the announcement, is that the platform currently hosts more than 6,000 abliterated models — models with their refusal behavior surgically removed. That is the scale problem this partnership is aimed at: not a hypothetical future risk, but a running total of already-published finetunes with guardrails stripped out, on the largest open-weight hosting platform there is. For what abliteration actually does to a model and what it costs in accuracy, see uncensored AI models: what abliteration actually does — this partnership is effectively an industry response to the exact gap that guide measures.
What's real here, and what isn't yet
The funding and the parties are real and verifiable: Baseten closed a $1.5B Series F in June 2026 at a $13B valuation, and Goodfire raised a $150M Series B earlier in 2026. Goodfire's own framing is a direct claim of responsibility: "Safety must be built into open models and provided by those who serve them." What's not yet real is any shipped artifact. The reporting is explicit that no specific technical implementation has been disclosed, and Baseten has only issued an open call for developer-ecosystem contributions. This is a funded, credible group of three companies stating an intention — worth tracking, not yet something to evaluate against a spec, because there isn't one.
Who should actually care right now
For the open-source model landscape this sits inside, see open-source AI tools ranked and every LLM in the directory.
Explore curated tools related to this guide: