AI Safety Partnership: Open Models and Business Risk

Base Labs, a research team launched by Baseten, has announced a new partnership focused on AI safety. Joining forces with Hugging Face and Goodfire, the group aims to develop and share methods for training and monitoring open-weight AI models. Open-weight models, which allow users to inspect and modify the underlying algorithms, are gaining traction as businesses demand transparency from their AI tools.

This collaboration puts a spotlight on how companies manage the inherent risks of deploying advanced AI, especially when the code isn’t locked behind proprietary walls. Business leaders now face a clear decision: how much control and visibility do they want over their AI systems, and what oversight mechanisms should they demand from partners and vendors?

For executives, the rise of open-weight models presents both opportunity and challenge. On one hand, open models offer greater transparency, enabling technical teams to audit and tailor AI to specific business needs. On the other, the burden of safety and compliance falls squarely on the company that chooses to deploy these tools.

The Base Labs partnership is set to publish new frameworks for monitoring open models—an essential step for businesses weighing the adoption of non-proprietary AI. This means that companies can’t just rely on vendor assurances. They must decide if their in-house expertise is up to the task of maintaining AI safety, or if they’ll depend on emerging standards led by this alliance.

Hugging Face’s involvement in this partnership signals a shift in how the industry thinks about trust in AI vendors. Traditionally, businesses betting on closed-source AI put their faith in the vendor’s internal safety practices. Open-weight models flip this script, making safety protocols visible, but also making the company responsible for implementing them.

Goodfire’s participation adds another layer: monitoring. As businesses bring open models in-house, they’ll need tools and processes to ensure these systems behave as expected—before and after deployment. Business leaders must decide: will they build internal teams for continuous monitoring, or invest in third-party tools and partnerships to shoulder this responsibility?

The partnership’s focus on publishing methods for training and monitoring open models is more than an academic exercise. This is a direct response to growing regulatory and reputational pressures. For any business leader considering open-weight AI, the question isn’t simply ‘Can we use this model?’ but ‘How will we manage it safely and prove compliance if regulators come knocking?’

Ignoring these questions courts unnecessary risk. This new alliance sets the stage for a future where open-weight model governance is a board-level conversation, not just a technical one.