.png)
By Kunal Tangri, Farsight · Adrian Lepers, Hugging Face · Jeff Boudier, Hugging Face
Until recently, financial firms had two good reasons to rule out open models: they weren't good enough, and the ones that came close were too expensive to run. That case no longer holds. Open models now meet the bar for demanding financial work, including multi-step agentic tasks, and they run efficiently enough that a firm can serve them on hardware it already owns or could reasonably buy.
So the constraint has moved. What decides how far AI gets inside a financial institution today is less how good the model is and more where it's allowed to run.
For that to work, open models have to make it through a firm's governance process, and there has to be an application that makes the effort worthwhile. Hugging Face sees the challenge from the model side via access, approval and governance. Farsight sees it from the application side, running inside firms' own environments and putting those models to work on pitch books, memos, and financial models. That's why we wrote this together.
Why financial institutions want the model inside their walls
When financial institutions talk to Farsight about running models in their own environment, they tend to give the same two reasons.
The first is data. The work that makes AI worth paying for at a financial firm runs on its most sensitive information. That material is what lets a system produce work that reflects how a particular firm thinks, instead of a competent generic draft. It's also what confidentiality obligations and data-residency rules keep from leaving the institution. A model hosted by an outside provider can be very capable and still never be allowed to see it, so the only way to use that data is to run the model inside the firm.
The second is control. Financial institutions, particularly in Europe, are watching how regulation and politics could change which providers and jurisdictions they can depend on, and which models stay available to them at all. Pricing is set by the provider, and a firm that has built its workflows around one API has little leverage when the price moves. Security is the newest worry. AI has made attacks cheaper and faster to mount, and each external dependency adds a route for data to leave the firm and a vendor whose security it has to take on trust. Models running internally sit behind the same controls as the rest of the firm's systems.
Wanting the model inside still doesn't get the budget approved. In our experience, it gets funded when an application people already use every day is creating the demand. Once that's true, the question changes from whether to try open models to how fast the firm can get them approved.
Bringing open models into the firm’s governance framework
For financial institutions, putting open models to work starts with a practical question: how do teams access the models they need while meeting the firm’s security and governance requirements? Hugging Face Enterprise Plus provides the controls to do that, connecting the open AI ecosystem to the identity systems, access policies and approval processes the firm already uses.
It starts with company-managed identities. With Managed SSO, employees access Hugging Face through the firm’s own identity provider, which controls their accounts from creation to deactivation. Enterprise Managed Users keep their work within the organization: these accounts cannot publish models or datasets to a personal profile or join another organization. The firm can give its teams access to Hugging Face while retaining control over their accounts and collaboration.
The next step is deciding which resources teams can use. Enterprise Plus Network Security lets firms require authenticated access from their corporate network and define policies for models, datasets and applications. A firm can make only approved resources available, or allow broader access while restricting specific repositories or publishers.
These policies can fit directly into existing technical, security and legal reviews. For example, a firm can let engineers browse model descriptions while restricting downloads to approved models. When a new model clears review, the firm can update its access policy through the Hub API. Teams have a clear path from discovering a useful model to requesting and gaining access to it.
Governance also extends to the work teams create. Resource Groups let firms assign access to private repositories by team or project, while token management gives administrators control over programmatic access. Audit Logs record changes to membership, repositories and security settings, giving the firm visibility into who changed what and when.
“Good governance should give teams a clear path to put open models to work. We built Enterprise Plus so companies can use Hugging Face with the security and accountability they expect from enterprise software, while giving their builders access to the innovation happening across open AI,” says Jeff Boudier, Hugging Face.
That creates a practical foundation for applications like Farsight. Hugging Face Enterprise Plus helps financial institutions govern access to open models, and Farsight puts those models to work on pitch books, memos and financial models inside the firm’s own environment. Together, they help connect model access and approval to the everyday work that makes AI valuable, with the firm retaining control over its data and deployment.
