Tech Giants Partner to Make Open AI Models Safer

Three tech companies are joining forces to solve one of artificial intelligence's biggest safety challenges, and they're doing it by making their work completely transparent.

Baseten's Base Labs research arm announced Wednesday it's partnering with Hugging Face and Goodfire AI to create new safety standards for open-weight AI models. These are AI systems anyone can download and modify, which makes them powerful tools for innovation but also vulnerable to misuse.

The timing couldn't be more critical. A technique called "abliteration" allows people to strip safety guardrails from AI models, and Hugging Face currently hosts over 6,000 models that have been modified this way. Instead of locking down the technology, these companies are betting on a different approach: building safety directly into how models work from the ground up.

"We believe openness to be an advantage for AI safety," Baseten explained. When AI systems are transparent, researchers can actually see how they make decisions and create better safeguards than they could with closed, secret systems.

Goodfire AI brings expertise in cracking open AI's "black box" to understand how models think. Baseten, fresh off raising $1.5 billion in June, provides the infrastructure to run AI models safely. Hugging Face hosts the massive library of open AI models used by developers worldwide.

The companies are calling their framework a new "standard" for the industry, one that's transparent and woven into the fabric of AI development rather than added as an afterthought. They're inviting the entire developer community to contribute ideas and improvements.

The Ripple Effect

This partnership represents a philosophical shift in how the tech industry thinks about AI safety. Rather than choosing between innovation and security, these companies are proving you can have both by working in the open.

The approach could influence how AI safety is handled across the entire industry. When major players with billions in funding commit to transparent safety standards, it sets expectations for everyone building AI tools.

By making their methods public and collaborative, they're also democratizing AI safety research. Smaller companies and independent researchers will be able to use and improve these tools, creating a rising tide that lifts all boats.

The message is clear: AI safety works best when everyone can see it, test it, and make it better together.