AI on devices you do not manage: what you can still control

Why do device controls stop at managed devices?

Every on-device control in this site's product notes reaches the laptop through something the company already controls. Bay and Noma Security state that they work through your existing EDR or MDM. Harmonic Security rolls out through Intune, JAMF, Kandji or Group Policy. Koi (Palo Alto Networks) uses its own agent or a Cortex XDR module, and Prompt Security (SentinelOne) uses its own browser extension and AI endpoint agent. All of those assume the device is enrolled.

On a personal laptop or a contractor's machine, none of that is there. You cannot push an agent, you cannot force a browser extension, and you cannot set the AI agent's managed settings. On the pages we read, none of the products we cover describes coverage for devices the company does not manage.

What can you still control?

You can control what the AI tools connect to. If a contractor's coding agent needs your source code or your internal APIs, those systems are yours, and access to them can require company identity, a managed device or a path through infrastructure you run.

• An AI or MCP gateway for the MCP servers and model access you provide, so calls to your systems pass a checkpoint you control.
• Access rules on the systems themselves: which identities and devices may reach source code, data stores and production.
• Company-provided AI tools in a cloud workspace, so the work happens on infrastructure you manage even when the screen is a personal laptop.
• A written policy that says which AI tools may be used with company data on personal devices, if any.

How should you think about contractors?

Contractors are the common mixed case. They often need the same access as employees and use their own laptops. The practical choices are to issue managed devices to the contractors with the most access, to route their work through a virtual desktop or cloud development environment, or to limit what their AI tools can reach on your side. Asking a contractor to install your security software on their own machine rarely works and raises its own privacy questions.

What does the finder do with this answer?

When you say devices are mostly unmanaged, the finder adds points to the browser extension, gateway and policy positions, and it shows a note that none of the products describes unmanaged-device coverage. When you say the estate is mixed, it adds weight to the browser and gateway positions for the unmanaged part. In our example profile of unmanaged devices with no clear view of AI use, the result is policy before a product.

What should you ask vendors anyway?

• Do you have any option for devices we do not manage, and what does it see?
• Can your gateway or proxy be required for access to our systems, so unmanaged devices have to pass through it?
• What do you see of AI use inside a virtual desktop or cloud development environment?

Sources