Building guardrails for GitHub Copilot cloud agent

Configure your enterprise so that Copilotクラウドエージェント will operate in a secure, compliant environment.

この記事で

Before you enable Copilotクラウドエージェント, it is good practice to set up your enterprise so you can be confident Copilot will operate within secure, predictable guardrails.

Learn about built-in protections

Copilotクラウドエージェント has a strong base of built-in security protections designed to protect against common risk points of AI agents. See GitHub Copilot クラウド エージェントのリスクと軽減策.

Plan policy settings

Plan your policies for Copilotクラウドエージェント in advance. Policies allow you to set a baseline for restrictions at the enterprise level, which organization owners can restrict further if needed.

Some questions to ask are:

Which policies don't apply?

The following Copilot policies don't apply to Copilotクラウドエージェント:

  • Content exclusions
  • Custom models (providing your own LLM API keys)
  • Private MCP registries

Adapt rulesets

Copilotクラウドエージェント is already restricted from actions like pushing to a default branch or merging pull requests. You can build on these default protections in branch rulesets. Copilotクラウドエージェント is subject to rulesets just like human developers.

To adapt your rulesets for Copilotクラウドエージェント:

  • Consider whether additional rules are required in repositories where agents will operate, such as requiring results from code scanning or Code Quality. If you have identified the organizations or repositories where Copilotクラウドエージェント will be enabled, you can apply a custom property to them so they're easy to target in a ruleset.
  • Consider whether Copilotクラウドエージェント will be blocked by any of your existing rulesets. Copilot can sign its commits, but it may not be able to follow other rules that restrict commit metadata.
  • Protect important Copilot and MCP configuration files with a CODEOWNERS file, and enable the "Require review from Code Owners" rule, so that edits to these files must be approved by specific teams. For filepaths to target, see Copilotカスタマイズ チートシート.

Set up your GitHub Actions environment

Copilotクラウドエージェント operates on GitHub Actions runners. Set up your runners and policies so that Copilot operates securely.

Store data and secrets

Continue to store data and tokens that you don't want Copilot to access as GitHub Actions variables or secrets. Copilot won't be able to access these in its sessions or environment setup steps.

If you need to provide data and secrets that Copilotクラウドエージェント does need, you'll be able to do this in a specific copilot environment.

Configure runners

Decide which runners you will use for Copilotクラウドエージェント. We recommend using GitHub-hosted runners, so that each Copilotクラウドエージェント runs on a fresh virtual machine. If you use self-hosted runners, we recommend using ephemeral runners.

Organization owners can restrict the Copilotクラウドエージェント's runners to a specific runner label, to be used automatically in all repositories. See GitHub Copilotクラウドエージェント用ランナーを組織に構成する方法.

Configure workflow policies

Decide whether GitHub Actions workflows should be blocked from running in pull requests that Copilotクラウドエージェント creates. See GitHub Copilotクラウドエージェントの設定を構成する.

By default, workflows are blocked from running until someone with write access approves them. Repository administrators will be able to disable this feature, so communicate with them in advance about your preferred setting.

Review default permissions

Review the default permissions for the GITHUB_TOKEN in your enterprise. See 企業でGitHub Actionsのポリシーを適用する.

This policy does not affect the token that Copilot will receive for its sessions, but the GITHUB_TOKEN is used in environment setup steps defined in copilot-setup-steps.yml workflow files.

Bear in mind that developers will be able to set their own permissions in these workflow files, and you should encourage them to use the minimum required permissions in all workflows.

Next steps

When you're ready to enable Copilotクラウドエージェント, see 企業GitHub Copilotクラウド エージェントの管理.