Product Guides are now only available at docs.box.com. Check out all the details on what's changed.

Box Status
Print

Choose your own model for the Box Agent

Rolling out/順次リリース中
Rolling out
We're excited to announce new enhancements to model controls and compliance management for Box AI, bringing admin-level governance, compliance alignment, and end-user model choice to power enterprise AI workflows.
 

What's New

Administrators will gain direct visibility and granular governance over AI models across their organization.

Key updates include:

  • Admin Model Governance: Admins will have granular control over how models are used across the enterprise, including the ability to disable/enable specific models based on compliance, tier, provider or preferences.
  • Admin Default Model Configuration: Administrators can configure the default model used for the Box Agent across Box AI Home, Notes, Files and Hubs.
  • Choose Your Own Model for the Box Agent (for Enterprise Advanced customers): End users can now, for the first time ever in Box, easily select from administrator-approved models to power their prompts that leverage the Box Agent.
  • Box AI Home for Regulated Environments: With administrator model selection, Box AI Home can now be enabled for organizations requiring FedRAMP Class D (High) and DoD Impact Level 4 (IL-4) compliance.

These governance settings apply directly to experiences powered by the Box Agent across Box AI Home, Notes, Files, Hubs, and agents used in Box AI Studio.

Key Capabilities & Admin Controls

Admin Console Model Governance
  • Granular control over models used across the organization: From the Admin Console Models page, admins will be able to see:
    • The models that support Pro mode
    • The compliance levels of that model (such as:  FedRAMP Class C (FRC / Moderate), FedRAMP Class D (FRD /High), DoD Impact Level 2 (IL2), and DoD Impact Level 4 (IL4))
    • Model pricing tier
    • The ability to actually disable/enable models
  • Impact on Custom Agents & Pro Mode: If a model powering a custom agent in AI Studio is disabled, that agent is deactivated until reconfigured with an approved model. Pro Mode becomes unavailable if all Pro-Mode-compatible models are disabled.
  • Default Activation State: Box AI Home remains off by default for everyone who has not previously enabled it, allowing administrators to configure approved models prior to rollout.

User Model Selection: Choose Your Own Model

  • Within Box AI Home, Notes, Files and Hubs, end users on Enterprise Advanced can choose from the dropdown of administrator-approved models when querying using the Box Agent. This means Box users can leverage the model of their choice for the task at hand - making it even easier to work across models and leverage AI right where your content lives.

Why This Matters

  • For Knowledge Workers & End Users: Puts the power of choice directly into employees' hands. For the first time, users on Enterprise Advanced can seamlessly switch between industry-leading models right inside Box AI Home, and anywhere the Box Agent can be used, empowering teams to match the best, most capable (and admin-approved) LLM to their specific task, prompt, or workflow for optimal results.
  • For Enterprise IT & AI Administrators: Delivers centralized, granular governance without compromising user productivity. Admins gain full visibility to curate, enable, or disable specific AI models, set organizational defaults, and ensure enterprise AI usage aligns with internal standards and budget tiers across all Box Agent experiences.
  • For Regulated Industries & Public Sector Teams: With admin-level model filtering, organizations adhering to stringent compliance standards—including FedRAMP Class D (High), FedRAMP Class C (Moderate), DoD IL2, and DoD IL4—can now confidently unlock Box AI Home and intelligent agents while staying strictly compliant.
Important Admin considerations:
  • Mobile Experience: This experience is coming soon on Mobile.
Top use cases unlocked:
  • Match the Right Model to the Right Task: Users on Enterprise Advanced have the agility to toggle between industry-leading LLMs directly in Box AI Home, Notes, Files and Hubs, pairing high-reasoning models with complex data synthesis and nimble models with rapid drafting.
  • Centralize AI Governance Without Slowing Innovation: IT and security teams can now curate approved foundation models, establish enterprise-wide defaults, and optimize AI spend across departments without friction.
  • Unlock Enterprise AI in High-Security & Regulated Environments: Confidently roll out Box AI Home and agent workflows across Federal, Defense, and public sector teams with the ability to select only FedRAMP Class D (High) and DoD IL4 models.
  • Future-Proof Custom Agents & Automated Workflows: Easily adapt and power tailored Box AI Studio agents with vetted, compliant models that meet organizational baselines.
  • Scale Intelligent Workflows and Automation: For organizations who can now adopt Box AI Home under these new granular compliance controls, can now leverage Box AI queries and agents across large sets of files, folders, hubs, and collections, as well as leveraging custom AI agents built in Box AI Studio within Box Automate workflows.
  • Foundational commitments
    • Zero Training on Customer Content: Enterprise files are never used to train or improve Box or third-party AI models.
    • Native Permissions Enforced: Box AI strictly respects existing Box permission architectures and access controls.

A Note on Release Timing:
This feature will be rolled out to customers from 8/24-9/4.

:point_right: Learn more about available models and compliance levels: box.com/models.
:point_right: See step-by-step instructions for leverage these new features here: https://docs.box.com/en/box-ai/admins/enabling-box-ai-studio-and-managing-agents

 

Related articles