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

Baseten is trying to turn open-model safety from an add-on into a default. On Wednesday, the company said its Base Labs research arm is launching a new safety infrastructure standard with Hugging Face and Goodfire AI, aimed at building evaluation and monitoring tools for open-weight models.

The timing is no accident. Open-weight models have become a battleground because they can be made more dangerous when their safeguards are removed, a practice the industry now calls abliteration. Hugging Face, which hosts open-source AI models, says it currently lists more than 6,000 abliterated models. That is a lot of bad homework floating around.

Base Labs, which Baseten spun up earlier this year, says it will develop and publish methods for training and monitoring open models. The company is pitching the effort as a standard for open models that is transparent and built into training and deployment, instead of something bolted on later when the damage is already done.

The technical details are still thin. Baseten and its partners haven’t said exactly how the system will work, but Goodfire’s role is easy enough to guess from its own pitch: it specializes in making AI less of a black box and more explainable. Baseten says openness is an advantage for safety because it gives more visibility into model behavior and more ways to turn safety research into practical controls than closed-source systems allow.

The money side is not exactly a shoestring effort either. Baseten raised a $1.5 billion Series F in June, which pushed its valuation to $13 billion. Goodfire raised a $150 million Series B earlier this year, led by B Capital. Now Baseten is opening the door to outside developers to help shape the framework. That’s the right instinct. Safety standards that live only in press releases are just vibes with a logo.