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

Base Labs, the research group Baseten spun up earlier this year, is launching an open-weight AI safety partnership with Hugging Face and Goodfire, TechCrunch reported on September 17. The collaboration aims to develop and publish methods for training and monitoring open models, putting safety tooling directly into the pipeline where open weights are built and shipped.

According to the report, the outputs are meant to be public methods rather than a safety capability wrapped inside a closed product. That distinction is the point: the work is intended to give the whole open ecosystem something reusable rather than a single vendor's advantage.

The partner list signals the intended scope. Hugging Face is the primary host and distribution channel for open models, while Goodfire works on interpretability and control of model internals. Together with Base Labs, the three cover distribution, training and monitoring.

Open-weight safety has been a long-running argument: once weights are released they cannot be recalled, so the community leans on alignment methods during training and on behavioral monitoring after release. Standardizing and publishing those methods is one practical way for the open ecosystem to answer regulatory and public pressure.

Base Labs sits inside Baseten, an inference and deployment infrastructure company, which suggests the research could reach real deployment tooling faster than a purely academic effort would.

The report describes direction and intent rather than deliverables: no timeline, first outputs or model coverage were disclosed. What to watch is whether reusable evaluation, training and monitoring tools actually ship, and whether outside researchers can independently verify them.