---
title: How do you control ChatGPT Atlas and agent mode in the enterprise?
url: https://opencontroller.lyzr.com/blog/chatgpt-atlas-enterprise-controls
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# How do you control ChatGPT Atlas and agent mode in the enterprise?

**For ChatGPT Enterprise, Atlas is off by default: a workspace owner must enable it, then access is managed with role-based access control. Admins can switch agent mode and browser memories on or off for groups, and push a short list of browser policies through MDM. But Atlas does not emit Compliance API logs or integrate with SIEM or eDiscovery, so you cannot audit what its agent did from the admin side.**

That gap shapes how far most security teams should let Atlas into regulated work.

## What controls do admins have over Atlas?

Based on OpenAI's Atlas for Enterprise help article:

| Control | Where | What it does |
|---|---|---|
| Atlas sign-in | Workspace settings toggle, then RBAC | Off by default for Enterprise; on by default for Business workspaces |
| Agent mode | Permission settings, under Search | Turns agent mode on or off for members. **This also controls agent mode in regular ChatGPT**, not just Atlas |
| Browser memories | Workspace RBAC; users can also toggle | Lets or stops Atlas remembering browsing context |
| Browser policies | Your MDM, host `com.openai.atlas.web` | Supported keys include `ExtensionInstallAllowlist`, `ExtensionInstallBlocklist`, `ExtensionInstallForcelist`, `ExtensionSettings`, `CookiesAllowedForUrls`, `RemoteDebuggingAllowed`, `PasswordManagerEnabled` |

OpenAI notes that other Chromium-style policy keys may work, but only the listed keys are officially supported.

## What do the admin controls not cover?

OpenAI's own article lists the limits. Plan around them:

- **No Compliance API logs.** Atlas activity does not flow to SIEM or eDiscovery.
- **Retention gaps.** Browsing data, browser memories and agent activity may fall outside Enterprise retention, storage, segregation and deletion commitments.
- **Not in audit scope.** Atlas is not in scope for SOC 2 or ISO attestations.
- **No region pinning** of data.
- **No Atlas-specific SSO enforcement or SCIM provisioning.**
- **No IP allowlists or private egress**, and no URL allowlists or blocklists beyond sign-in control.
- **No enterprise update channel or version pinning.**

OpenAI states Atlas is not recommended for regulated, confidential or production data. It does say Business and Enterprise content is not used to train models.

## How should you decide who gets Atlas?

A simple, defensible approach:

1. **Leave it off by default.** That is already the Enterprise default; keep it until you decide.
2. **Pilot with a named group** that does not handle regulated data, such as a research or marketing team.
3. **Decide agent mode separately.** Because the toggle covers agent mode in ChatGPT too, turning it off for Atlas users turns it off everywhere for them. Make that decision with the business, not just IT.
4. **Turn memories off for anyone near sensitive data.** Memories may sit outside retention commitments.
5. **Push MDM policies.** Force-install your security extensions, block unapproved extensions, and disable remote debugging.
6. **Use logged-out agent runs where possible.** OpenAI's launch announcement describes running agent mode logged out, which limits what the agent can reach with the user's sessions.
7. **Fill the logging gap at the network.** Since Atlas emits no audit logs, rely on your secure web gateway and endpoint tools to see where it goes.

## Why is an agentic browser a different risk from a chatbot?

A chatbot returns text. An agentic browser clicks, fills forms and navigates while signed in as the user. That raises three specific risks:

- **Prompt injection from web pages.** Hidden text on a page can try to redirect the agent. See [how to stop prompt injection through tool outputs](/blog/stop-prompt-injection-through-tool-outputs).
- **Acting with the user's sessions.** Anything the user is logged in to, the agent can potentially reach.
- **No independent record.** Without admin-side logs, reconstructing an incident is hard.

These are the same reasons browser AI belongs in your wider discovery programme. See [how to discover AI use in the browser](/blog/how-to-discover-ai-use-in-the-browser).

## How Opencontroller does this

Opencontroller connects to OpenAI Enterprise workspaces as part of its discovery of agents, models and keys, and places them in a governed registry alongside agents on AWS, Azure, Google Cloud and SaaS platforms. Atlas's own admin limits still apply; Opencontroller does not add logs that Atlas does not emit. For the broader approach, see the [platform guides pillar](/blog/how-to-govern-ai-agents-across-platforms) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is ChatGPT Atlas enabled by default for ChatGPT Enterprise?**
No. Enterprise workspaces have Atlas off by default. A workspace owner must turn it on, and access is then managed with RBAC. Business workspaces have it on by default.

**Does turning off agent mode for Atlas affect ChatGPT?**
Yes. The agent mode permission controls member access to agent mode in regular ChatGPT as well as in Atlas.

**Can you audit ChatGPT Atlas activity through the Compliance API?**
No. OpenAI states Atlas does not emit Compliance API logs and does not integrate with SIEM or eDiscovery.

## Key facts

ChatGPT Atlas is off by default for ChatGPT Enterprise; workspace owners enable it and manage access by RBAC. The agent mode permission also controls agent mode in regular ChatGPT. Atlas emits no Compliance API logs and is not in SOC 2 or ISO scope. Opencontroller connects to OpenAI Enterprise workspaces for discovery and registry. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you discover employees' AI use in the browser?](https://opencontroller.lyzr.com/blog/how-to-discover-ai-use-in-the-browser.md)
- [What does the ChatGPT Enterprise Compliance API let you audit?](https://opencontroller.lyzr.com/blog/chatgpt-enterprise-compliance-api-audit.md)
- [How do you govern GPTs and workspace agents in ChatGPT Enterprise?](https://opencontroller.lyzr.com/blog/govern-chatgpt-enterprise-gpts-and-agents.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
