Skip to content
Endpoint AI Finder

FIVE CONTROL TYPES

AI endpoint control types: on-device, browser, network, gateway and policy only

Short answer

AI controls for employees sit in five places: on the device, in the browser, on the network path, at an AI or MCP gateway, or in policy with no product. Each sees a different slice: a browser extension sees chat tools but not a coding agent in a terminal, and a network proxy sees traffic that leaves the laptop but not a local MCP server. The useful question is which slice your risk lives in, and the finder answers it from your situation.

By The Device Desk · Endpoint AI Finder · Published 2026-10-11

Desk research from public vendor pages read on 27 September and 11 October 2026. Profiles follow our published method; we did not run the products.

Vendor pages read September and October 2026 · updated 2026-10-13

What does each control type see?

Where AI controls can sit around one employee device: on the device, in the browser, on the network, at an AI or MCP gateway, or in policy.AI servicesMCP serversDEVOn-device controlBRWBrowser extensionNETNetwork or SSE proxyGWYAI or MCP gatewayPOLPolicy only
The map shows where each product's own pages describe a control, or where your answers point. It is not a measure of how well a product works.
  • DEV

    On-device control

    Acts on the laptop itself, where agents run commands and read files: a resident agent, a binary pushed through EDR or device management, or hooks inside the agent.

    Described by 8 of 19 products, partly by 8

    Read the DEV guide
  • BRW

    Browser extension

    Sits in the browser and watches or controls what employees type, paste and upload into AI chat tools.

    Described by 6 of 19 products, partly by 2

    Read the BRW guide
  • NET

    Network or SSE proxy

    Inspects web traffic to AI services as it leaves the device, through a proxy or a security service edge.

    Described by 3 of 19 products, partly by 0

    Read the NET guide
  • GWY

    AI or MCP gateway

    Routes AI and MCP traffic through one checkpoint that can filter, block and log calls to models and tools.

    Described by 5 of 19 products, partly by 1

    Read the GWY guide
  • POL

    Policy only

    A written policy, an approved-tools list and the settings AI tools already offer, without buying a dedicated product.

    No product needed

    Read the POL guide

Which product describes which control type?

From each product's own pages, read 27 September or 11 October 2026. Not in our sources means we did not find it; it may exist.
ProductMapOn-device controlBrowser extensionNetwork or SSE proxyAI or MCP gateway
AI Access Security (Palo Alto Networks)
AI Access Security: described at 1 of 4 product control positions (network or sse proxy).AI servicesMCP serversDEVBRWNETGWYPOL
Not in our sourcesNot in our sourcesDescribedNot in our sources
Akamai Workforce Protector
Workforce Protector: described at 1 of 4 product control positions (browser extension), partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyDescribedNot in our sourcesNot in our sources
Bay
Bay: described at 1 of 4 product control positions (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedNot in our sourcesNot in our sourcesNot in our sources
Bloom Security
Bloom: described at 0 of 4 product control positions, partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyNot in our sourcesNot in our sourcesNot in our sources
Cloudflare AI Gateway
Cloudflare AI Gateway: described at 1 of 4 product control positions (ai or mcp gateway).AI servicesMCP serversDEVBRWNETGWYPOL
Not in our sourcesNot in our sourcesNot in our sourcesDescribed
CrowdStrike Falcon Guardian
Falcon Guardian: described at 1 of 4 product control positions (on-device control), partly at 2 (browser extension, ai or mcp gateway).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedPartlyNot in our sourcesPartly
Ent
Ent: described at 2 of 4 product control positions (on-device control, browser extension).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedDescribedNot in our sourcesNot in our sources
Glow Security
Glow: described at 0 of 4 product control positions, partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyNot in our sourcesNot in our sourcesNot in our sources
Harmonic Security
Harmonic: described at 2 of 4 product control positions (browser extension, ai or mcp gateway), partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyDescribedNot in our sourcesDescribed
Koi (Palo Alto Networks)
Koi: described at 1 of 4 product control positions (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedNot in our sourcesNot in our sourcesNot in our sources
Lasso Security
Lasso: described at 3 of 4 product control positions (browser extension, network or sse proxy, ai or mcp gateway), partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyDescribedDescribedDescribed
Neo
Neo: described at 1 of 4 product control positions (on-device control), partly at 1 (browser extension).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedPartlyNot in our sourcesNot in our sources
Noma Security
Noma: described at 1 of 4 product control positions (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedNot in our sourcesNot in our sourcesNot in our sources
Onyx Security
Onyx: described at 3 of 4 product control positions (on-device control, browser extension, ai or mcp gateway).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedDescribedNot in our sourcesDescribed
Pluto Security
Pluto: described at 0 of 4 product control positions, partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyNot in our sourcesNot in our sourcesNot in our sources
Prompt Security (SentinelOne)
Prompt Security: described at 2 of 4 product control positions (on-device control, browser extension).AI servicesMCP serversDEVBRWNETGWYPOL
DescribedDescribedNot in our sourcesNot in our sources
Sun Security
Sun: described at 0 of 4 product control positions, partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyNot in our sourcesNot in our sourcesNot in our sources
Zenity
Zenity: described at 1 of 4 product control positions (ai or mcp gateway), partly at 1 (on-device control).AI servicesMCP serversDEVBRWNETGWYPOL
PartlyNot in our sourcesNot in our sourcesDescribed
Zscaler Generative AI Security
Zscaler: described at 1 of 4 product control positions (network or sse proxy).AI servicesMCP serversDEVBRWNETGWYPOL
Not in our sourcesNot in our sourcesDescribedNot in our sources
Products describing it8635

How do the control types fit together?

The five types are not rivals. They sit at different points between the person, the AI tool and the systems the tool reaches. The closer a control sits to the action, the more it can see: an on-device control can see a coding agent read a credentials file, which nothing on the network will ever see. The further out a control sits, the less you need to install: a gateway or a proxy works without touching the device, as long as the traffic passes through it.

That is why the device question matters so much in the finder. On managed devices, an on-device control or a browser extension can be rolled out through tools you already run. On devices you do not manage, the options shrink to what the traffic passes through, and to policy.

Policy is the fifth type because it is the cheapest and it never goes away. Even companies that buy a product still need an approved-tools list, a rule for who may connect which MCP servers, and someone who reviews new requests.

Questions people ask

Can one product cover all five control types?

Not on the pages we read. No product here describes all four product types, and policy is always your own work. Many companies combine a device or browser control with a gateway.

Which control type sees the most of coding agents?

On-device control, because coding agents act on the device: they read files, run shell commands and start local MCP servers that never touch the network. A gateway sees only the calls routed through it.

Why is policy only a control type?

Because for some companies it is the right first step: few users, general information, or devices you cannot install anything on. The finder says so when your answers point there.