How to Control Which AI Apps Are Allowed on Company Laptops

To manage the risks of shadow AI, IT and security teams are turning to endpoint governance solutions. A combination of a central AI gateway like Bifrost and an endpoint agent like Bifrost Edge gives organizations visibility and control over the specific AI applications employees can use on company-issued devices.

The adoption of AI tools in the workplace has moved from a niche experiment to a daily reality. Employees are using AI to draft emails, summarize documents, and write code, often reaching for the most convenient tool for the job, regardless of official approval. This unsanctioned use of AI applications is known as "shadow AI," and it creates significant blind spots for security and compliance teams. When employees use unvetted tools, they can inadvertently expose sensitive corporate data, violate compliance regulations like GDPR, and introduce unreliable or biased outputs into business workflows.

Simply banning all AI is not a viable option, as it stifles the productivity gains these tools provide. A more effective approach is to govern AI usage directly on the endpoint: the company-issued laptops where this activity happens. This involves discovering which applications are being used, setting clear policies on which are permitted, and enforcing those rules on every machine. Modern AI governance platforms accomplish this with a two-part architecture: a central gateway for policy management and a lightweight endpoint agent for enforcement.

The Challenge: You Can't Govern AI You Can't See

Before any policy can be enforced, you need to know what to enforce it on. The first step in controlling AI applications is gaining visibility. Most organizations have no real-time inventory of the AI desktop apps, browser-based tools, and coding agents running on their fleet. Employees are often three times more likely to be using generative AI than company leaders realize.

An endpoint governance agent solves this visibility problem. It runs on each employee's machine (macOS, Windows, and Linux) and inventories the AI applications and services in use. This data is sent back to a central dashboard, giving administrators a live, fleet-wide catalog of every AI tool being used, by whom, and how often. This turns guesswork into a concrete dataset, forming the foundation for an effective governance strategy.

Step 1: Discover and Inventory Every AI Application

The process begins with deploying a lightweight agent to every company laptop. This is typically handled silently through existing Mobile Device Management (MDM) solutions like Jamf, Microsoft Intune, or Kandji, requiring no action from the end-user.

Once installed, the agent identifies AI-related traffic and application usage on the device. It covers the primary ways employees use AI today:

Desktop AI Apps: Standalone applications like Claude Desktop or the ChatGPT app.
AI in the Browser: Web-based services such as chatgpt.com or claude.ai.
Coding Agents: Tools used by developers in the terminal and IDE.

This discovery process is continuous. When a new AI application appears on any device in the fleet, it's automatically added to the central inventory for review.

Step 2: Set Centralized Allow/Deny Policies

With a complete inventory, administrators can move from visibility to control. In a central management console, such as the one provided by the Bifrost AI gateway, every discovered application can be reviewed and assigned a policy.

The workflow is straightforward:

Review Discovered Apps: The dashboard shows a list of all AI tools found across the fleet, such as "ChatGPT (Desktop)" or "Claude Code."
Approve or Deny: For each application, an administrator can set its status to "Approved" or "Denied."
Deploy Policy: The decision is saved as a central policy.

This approval workflow allows for a granular approach. Instead of a blanket ban, teams can sanction the use of specific, vetted applications that meet their security and compliance standards while blocking those that do not. For an added layer of control, policies can also be applied to the MCP servers that agentic AI tools connect to, preventing them from executing unapproved actions like arbitrary code execution or file system access.

Step 3: Enforce the Policy on Every Laptop

Once a policy is set in the central gateway, the endpoint agent on each laptop enforces it. The agent, like Bifrost Edge, transparently routes all AI traffic from the laptop through the central AI gateway. This ensures that every request is checked against the organization's policies before it proceeds.

The experience is designed to be seamless for the end-user:

Approved Apps Work Normally: When a user opens an approved application, it functions without any change in their workflow. The governance happens invisibly in the background.
Denied Apps Are Blocked: If a user tries to launch or use a denied application, the agent blocks the request. The user receives a clear notification on their device explaining that the application is not permitted by company policy.

This model allows organizations to enable productive AI use while maintaining control. It also ensures that all usage of approved tools is routed through the gateway, where other governance controls—such as budget limits, rate limits, and guardrails to prevent data leaks—are applied. The result is a secure and compliant AI ecosystem that doesn't hinder innovation.

Conclusion: From Shadow AI to Governed AI

Controlling which AI applications can run on company laptops is a critical step in managing the risks of shadow AI. By pairing a central policy engine with an endpoint enforcement agent, organizations can move from a state of zero visibility to one of complete control. This approach allows security teams to discover every AI tool in use, create clear allow/deny policies, and enforce them on every device in the fleet. It transforms AI from a source of unmanaged risk into a governed, secure, and productive tool for the entire organization.