Govern what AI models employees can access with Merge

On June 12, 2026, the US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals. 

Model governance is coming. Governments are becoming increasingly specific about which models may be accessed and used in a variety of contexts. Organizations have additional reasons to care: cost, security, and role-based requirements are important factors for model access across different teams. 

Until now, there has been no reliable way to enforce AI model access at the employee level. 

Merge has built model governance.

AI model governance in Agent Handler for Employees

Agent Handler now lets IT define model access rules based on attributes synced from your identity provider (e.g. Okta, Microsoft Entra) and HR system.

SCIM brings in group membership, role, department, employment type, and work location from your IdP. The HR system integration adds citizenship and work authorization. Those attributes can be used to configure access rules.

Rules can be built from any employee attribute Agent Handler syncs: citizenship status, groups, work location, employment type. 

When someone joins, their model access is set based on those attributes. If their work authorization restricts access to certain models, once that rule is configured and approved by an administrator, that policy is applied automatically. The same mechanism governs access by role and team, so an engineer in eng-platform gets a different catalog than a support rep. When an employee's status changes in the HRIS, Agent Handler picks it up without anyone touching a key.

Employees can see which models they have access to and use any subset of them. If they need something outside their current set, they can request short-term access. IT approves or denies, and temporary access expires on its own.

What comes next

The Fable 5 situation won't be the last time a government draws a line around a specific model.

Model-specific regulations are going to become more common. 

There’s no reason to expect this to change direction. For enterprises running AI at scale, model governance is going to be a hard requirement, and we’ve built the infrastructure for it at Merge. 

If you want to see how model governance works in Agent Handler, book a demo.