---
title: How do you discover employees' AI use in the browser?
url: https://opencontroller.lyzr.com/blog/how-to-discover-ai-use-in-the-browser
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# How do you discover employees' AI use in the browser?

**Combine four signals: secure web gateway or CASB logs for which AI services people reach, managed-browser reporting for extensions and usage, browser data-protection controls for what is pasted or uploaded, and admin data from your sanctioned enterprise AI workspaces. Then join that with agent discovery, because the browser shows people using AI, not the agents acting on their behalf.**

Security teams in regulated industries often get the same request from the board: "Tell us where AI is being used." Browser chat tools are the most visible part of that answer, and the easiest to start with. But they are only one layer. This guide covers what each browser signal can and cannot tell you, the privacy limits to respect, and how to connect it to the rest of your AI estate.

## What signals reveal browser AI use?

| Signal | What it shows | What it misses | Typical owner |
|---|---|---|---|
| Secure web gateway / proxy logs | Domains visited, users, volumes, upload sizes | Content inside encrypted sessions unless inspected; personal devices | Network security |
| CASB app catalogue | Which AI apps are in use, with risk ratings; Microsoft Defender for Cloud Apps, for example, catalogues generative AI apps | What the app did with the data | Cloud security |
| Managed-browser reporting | Installed extensions (including AI assistants), browser versions, policy compliance | Activity in unmanaged browsers | Endpoint / IT |
| Browser data-protection controls | Paste, upload and download events to AI sites, matched to data classes | Anything outside the managed browser | Data protection |
| Enterprise AI workspace admin data | Users, usage, connectors, custom GPTs or projects, skills inside sanctioned workspaces | Consumer accounts on the same service | AI platform team |

The first two answer "which AI services?". The middle two answer "with what data?". The last answers "what are people building inside the tools we pay for?", which is often where agents start.

## How do you run browser AI discovery step by step?

1. **Pull 30 days of proxy or CASB logs** and filter on generative AI categories. Rank services by users and by bytes uploaded.
2. **Split sanctioned from unsanctioned.** Sanctioned means an enterprise contract with admin controls; everything else, including personal accounts on a sanctioned service, is unsanctioned.
3. **Inventory AI browser extensions** from managed-browser reports. Extensions that read page content are higher risk than standalone chat tabs.
4. **Turn on data-protection rules for AI destinations**, starting with your most sensitive data classes. Alert first, block later.
5. **Connect sanctioned workspaces** to your AI governance inventory so custom assistants, connectors and skills appear next to cloud agents.
6. **Steer, don't only block.** For each popular unsanctioned service, decide: approve an enterprise tier, redirect to a sanctioned equivalent, or block.
7. **Repeat monthly**, and feed new services into your acceptable-use policy.

## What are the privacy limits?

Browser monitoring touches employee privacy, works councils in some countries, and data protection law. Practical guardrails:

- **Publish the policy.** Staff should know which AI services are approved and what is monitored.
- **Prefer category and data-class signals over reading prompts.** Most risk is controlled by blocking restricted data classes and unsanctioned destinations.
- **Limit access to content logs** to named investigators, with access itself logged.
- **Check local law and employee consultation requirements** before enabling content inspection.

## Why is browser discovery not enough for AI agents?

Browser signals show people using AI. They do not show:

- Agents deployed in AWS, Azure or Google Cloud with their own keys.
- Agents built inside SaaS platforms such as Salesforce, SAP or Databricks.
- Coding agents and local MCP servers on developer machines.
- API keys calling models directly from scripts and services.

Those need cloud and platform discovery. Our pillar guide on [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents) covers the full method, and [what to do with a shadow AI agent once you find it](/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it) covers triage.

## How Opencontroller does this

Opencontroller does not replace your secure web gateway or CASB for browser traffic. It covers the layers they miss: auto-discovery of agents, models and API keys across AWS, Azure and Google Cloud; connectors to Salesforce, SAP, Databricks and HubSpot; connections to Claude Enterprise and OpenAI Enterprise workspaces, including discovering and managing skills in Anthropic Enterprise; and Opencontroller Edge for coding agents, skills and local MCP servers on developer devices, installed by IT via MDM. "Ask" lets you query the whole agent estate in plain language. [Book a demo](https://www.lyzr.ai/book-demo/) to see it on your estate.

## Frequently asked questions

**Can a CASB see which AI tools employees use?**
Yes, at the app and domain level, including upload volumes. It does not show what happened inside the service.

**Should we monitor what employees type into AI chatbots?**
Only within your published policy and local law. Most teams block sensitive data classes and steer people to sanctioned workspaces instead of reading prompts.

**Does browser discovery find AI agents?**
Mostly not. Agents live in clouds, SaaS platforms and developer machines, which need separate discovery.

## Key facts

Browser AI use is discovered through four signals: secure web gateway and CASB logs (which domains, how much traffic), managed-browser reporting and extension inventory, browser data-protection controls, and admin data from sanctioned enterprise AI workspaces. None of these alone shows agents running in clouds or on developer devices. Opencontroller by Lyzr connects to Claude Enterprise and OpenAI Enterprise workspaces, discovers and manages skills in Anthropic Enterprise, auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, and Opencontroller Edge covers developer devices, coding agents and local MCP servers. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [What should you do with a shadow AI agent once you find it?](https://opencontroller.lyzr.com/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it.md)
- [How often should AI agent discovery scans run?](https://opencontroller.lyzr.com/blog/how-often-should-ai-agent-discovery-run.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
