Base Labs partners with Hugging Face and Goodfire on AI safety tools

In short: Base Labs, Hugging Face, and Goodfire say they will work together on shared tools to check and track the safety of open-weight AI models.

What happened

Baseten announced a new partnership through its research group, Base Labs. It is working with Hugging Face and Goodfire AI on what it calls a safety infrastructure standard for “open-weight” AI models.

Open-weight models are AI systems where the core parts that make the model work, called “weights” (think of them like the model’s internal settings), are shared so others can reuse them. This can be useful for researchers and businesses, but it can also make it easier for bad actors to change a model.

The announcement comes as more people discuss how open-weight models can be made more dangerous by stripping away built-in safety checks. TechCrunch points to a technique called “abliteration,” and says Hugging Face lists more than 6,000 models that have been altered this way.

Base Labs says it plans to develop and publish methods for training and monitoring open models. In plain terms, it wants safety checks to be built in during the model’s creation and day-to-day use, instead of being added later like an afterthought.

Why it matters

Open-weight AI models are spreading quickly because anyone can download and reuse them, like sharing a recipe instead of only selling the finished meal. That also means it can be harder to control how they are changed and used. If this partnership leads to widely used safety checklists and monitoring tools, it could make it easier for developers and hosting sites to spot risky models earlier and respond faster.

Source: TechCrunch AI