---
title: How do GitHub Copilot enterprise and organization policies interact?
url: https://opencontroller.lyzr.com/blog/github-copilot-enterprise-vs-organization-policies
published: 2026-11-15
updated: 2026-11-15
author: Lyzr
---

# How do GitHub Copilot enterprise and organization policies interact?

**Enterprise owners decide first. For most Copilot policies they can enable a feature, disable it, or choose "Let organizations decide", which hands the choice to each organization's owners. When a user holds Copilot seats from several organizations in one enterprise, most features follow the least restrictive setting, while a short list of sensitive ones follows the most restrictive.**

This trips up governance teams. A feature you disabled in one organization can still be on for a developer who also holds a seat elsewhere. Here is how the hierarchy works and what to audit.

## Where are Copilot policies set?

Copilot policies control which features, agents and models users can access. They generally apply to users on Copilot Business or Copilot Enterprise plans.

- **Enterprise level.** Owners manage policies on the enterprise's **AI controls** tab.
- **Organization level.** Owners manage policies in organization settings, for their own licensed users, wherever the enterprise has delegated the decision.
- **Enterprise-assigned users.** Users licensed directly by the enterprise, not through an organization, are not covered by "Let organizations decide". A separate "Policies for enterprise-assigned users" setting decides whether those policies default to enabled or disabled.

Two details matter for coding agents. Access to Copilot cloud agent is granted by the enterprise choosing exactly which organizations get it. And the Copilot app and Copilot CLI have their own client policies.

## What happens when policies conflict?

Conflicts arise when one person has seats from more than one organization. GitHub resolves them per feature.

| Rule | What it means | Example features (per GitHub docs) |
|---|---|---|
| Least restrictive | Enabled in any organization means enabled for the user | Copilot Chat in the IDE, agent mode in IDE chat, code review, Copilot CLI, MCP servers in Copilot, editor preview features, web search |
| Most restrictive | Disabled in any organization means disabled for the user | Suggestions matching public code, Copilot Metrics API, code review for members without a licence |
| All must opt in | Only on when every organization sets Enabled | Semantic indexing for non-GitHub repositories |

Across separate enterprises, the most restrictive policy usually applies. GitHub notes exceptions, such as paid AI credit usage, which applies per enterprise.

The practical consequence: if MCP servers in Copilot must be off, disabling them in one organization is not enough. Set the policy at enterprise level, or make sure no organization a user belongs to enables it.

## Which settings should stay at enterprise level?

Delegation is useful, but some decisions should not vary by team. A sensible split:

1. **Enforce at enterprise level:** MCP servers in Copilot, Copilot CLI, editor preview features, suggestions matching public code, and which organizations get cloud agent. These carry the most data and execution risk.
2. **Enforce or tightly delegate:** model availability. Our guide on [controlling which models GitHub Copilot can use](/blog/control-which-models-github-copilot-can-use) covers the trade-offs.
3. **Delegate:** features with low risk where teams differ in need, such as commit message generation.

Write the split down. An undocumented "Let organizations decide" is the most common source of drift.

## How do you audit Copilot policy drift?

Use this checklist quarterly, or after any reorganisation.

1. Export the enterprise AI controls settings and record which policies are enabled, disabled or delegated.
2. For each delegated policy, list every organization's value.
3. Find users with seats in more than one organization. Their effective setting follows the conflict rules above.
4. Check the "Policies for enterprise-assigned users" default.
5. Confirm which organizations have cloud agent access and whether that still matches intent.
6. Review audit log events for policy changes; see [GitHub Copilot audit log events](/blog/github-copilot-audit-log-events).
7. Compare MCP policy with what developers actually run; see [controlling MCP servers in GitHub Copilot](/blog/control-mcp-servers-in-github-copilot).

## Why isn't Copilot policy enough on its own?

Copilot policies govern Copilot. Developers also run other coding agents, local MCP servers and agent skills that no Copilot setting sees. Agent governance needs an inventory that spans tools, owners for each agent, and evidence that controls are applied. The [guide to governing AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms) explains the wider model.

## How Opencontroller does this

Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers, installed by IT via MDM. It complements Copilot's own policies rather than replacing them. Discovered tools and MCP servers feed the governed registry, and Posture flags risky permission and configuration patterns. To see this on your developer estate, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can an organization owner override an enterprise policy?**
Only when the enterprise has chosen "Let organizations decide". An explicit enterprise enable or disable applies to all organizations.

**Does disabling a feature in one organization turn it off for a user?**
Not always. For least-restrictive features, a seat in any organization with the feature enabled keeps it on.

**Do Copilot policies apply to personal Copilot plans?**
Mostly no. Policies generally apply to Business and Enterprise plan users, and a personal plan is cancelled when a user joins one of those plans.

## Key facts

GitHub Copilot policy hierarchy: enterprise owners set policies on the enterprise AI controls tab - enable, disable or "Let organizations decide"; Copilot cloud agent access is granted to specific organizations; enterprise-assigned users follow a separate "Policies for enterprise-assigned users" default. Multi-org users in one enterprise: least restrictive applies to most features (Chat in IDE, agent mode, code review, CLI, MCP servers in Copilot, editor preview features); most restrictive applies to sensitive ones (suggestions matching public code, Copilot Metrics API). Across enterprises the most restrictive usually applies. Opencontroller (Lyzr) Edge extends control to coding agents, skills and local MCP servers on developer devices. Pricing starts from $250K per deployment per year.

## Related

- [How do you control which MCP servers GitHub Copilot can use?](https://opencontroller.lyzr.com/blog/control-mcp-servers-in-github-copilot.md)
- [How do you control which models GitHub Copilot can use?](https://opencontroller.lyzr.com/blog/control-which-models-github-copilot-can-use.md)
- [Which GitHub audit log events show Copilot policy and agent activity?](https://opencontroller.lyzr.com/blog/github-copilot-audit-log-events.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
