Guide

Ways to Control AI Use on Business Devices

A written policy, a browser extension, a network filter, account connections, MDM settings or an endpoint agent: what each one sees, what it misses, and how much work it takes to run.

Last updated

The Short Answer

There are six common ways to control how staff use AI on business devices: a written AI use policy, a browser extension, a network or DNS filter, connections to the company's Microsoft 365 or Google Workspace accounts, MDM settings, and an endpoint agent. Each one sees a different part of AI use, so most organizations combine a policy with one or two technical controls.

The right mix depends on where staff use AI. If it is mostly chat in a browser on office computers, a narrow control may be enough. If staff also use desktop AI apps, coding tools, personal accounts or work from home, a control needs to see those places too.

Side by Side

ApproachWhat It SeesWhat It MissesWork to Run
Written AI Use PolicyNothing on its own. It sets the rules and records who accepted them.Whether anyone follows it.Low: write it once, collect sign-off, review it yearly.
Browser ExtensionAI sites visited in the browsers it is installed in, and often what is typed into them.Other browsers, desktop AI apps, command-line tools, MCP servers and local AI runtimes.Low to medium: push it with browser management and keep it from being removed.
Network or DNS FilterWhich AI services a device reaches while its traffic passes through the filter. Can block AI sites by name.Devices off that network or path, unless traffic is routed back. Seeing what was sent, or which account was used, requires decrypting traffic.Medium: depends on the filter and on decryption.
Account Connections (Microsoft 365, Google Workspace)AI apps staff have connected to their work accounts, from any device, and built-in AI settings in those suites.AI used without signing in with a work account, and what anyone types into an AI tool.Low: an administrator grants access once.
MDM SettingsWhich apps are installed on managed devices. Can block app installs and set browser policies.AI used in a browser, and what is typed or uploaded.Medium: policies per platform, kept up to date as AI apps change.
Endpoint AgentAI use on the computer itself: every browser, desktop AI apps, command-line tools, MCP servers and local AI runtimes, wherever the computer is. Can check prompts and uploads before they leave.Devices it is not installed on, such as phones and personal computers.Medium: install through an RMM or MDM, then tune rules per client.

Each Approach

A Written AI Use Policy

Every other control needs one. A policy says which AI tools are approved, which data must never go into them, and who to ask about a new tool. It changes behavior when staff actually read and accept it, so record who accepted which version. Our AI policy template is a starting point.

A Browser Extension

An extension is quick to roll out and sees AI chat in that browser well. Its limit is scope: it cannot see a second browser, a desktop AI app or a coding tool in a terminal, and staff can sometimes switch it off if browser management does not lock it.

A Network or DNS Filter

A filter can block AI sites across a whole office with one rule. It sees names, not content, unless it decrypts traffic, and it sees nothing from a laptop on home Wi-Fi unless that traffic is routed back through it.

Account Connections

Connecting to the company's Microsoft 365 or Google Workspace shows which AI apps staff have granted access to their work accounts, including from phones. It cannot see AI used on a personal account.

MDM Settings

MDM can stop an AI desktop app from being installed and set browser policies. Most AI use happens in a browser, so MDM alone leaves the largest part open.

An Endpoint Agent

An agent runs on each computer, so it sees AI use in browsers, desktop AI apps and command-line tools on any network, and can check data before it leaves the device. It needs installing and a sensible starting policy. Starting in a mode that only observes avoids surprising anyone.

Putting Them Together

A common setup for a small or mid-sized business is a written policy that staff accept, an endpoint agent on company computers, and account connections to cover phones and apps connected to work accounts. MDM handles app installs where it is already in place.

ProctoredAI combines three of these: policy acceptance, an endpoint agent for Windows, macOS and Linux (with a managed extension for Chromebooks), and Microsoft 365 and Google Workspace connections, run by an MSP from one console. See what it does, the shadow AI guide and how to evaluate AI governance tools.

Questions

What is the easiest way to start controlling AI use?

Write and share an AI use policy, then turn on a control in observe-only mode for two weeks to see which AI tools staff actually use. Decide what to approve and block from that list.

Is blocking all AI tools a good idea?

Blanket blocks tend to push AI use to personal phones and laptops, where nothing is visible. Approving one or two tools with work accounts and stopping sensitive data usually works better.

Can AI use be controlled on personal devices?

Only partly. Account connections show AI apps tied to work accounts from any device. A personal device with no work sign-in is outside every technical control, which is why the written policy matters.

Do we need to decrypt traffic?

A network filter needs to decrypt traffic to see what was sent. An endpoint agent can check prompts on the device itself, though some agents, ProctoredAI included, install a certificate on the device to do it.

See What Your Clients' Staff Actually Use

MSPs: deploy from your RMM in Passive mode (observe only) and review the results with each client. Businesses: ask your IT provider about ProctoredAI, or contact us and we will help find one.

Running your own IT? See ProctoredAI for your business