On July 27, 2026, GitHub announced that the GitHub Copilot app and Copilot cloud agent can now use enterprise managed settings. An enterprise can use one managed-settings.json file to govern multiple Copilot clients, bringing selected controls already used with Copilot CLI and VS Code to the desktop app and cloud-agent tasks.
The controls cover which plugins are available, which plugin marketplaces developers may install from, and whether Copilot can bypass approval prompts before it runs commands, accesses files, or fetches URLs. The documentation also lists an option to make auto model selection the default for new conversations. These settings define tool and model boundaries; they do not complete code security or data governance.
GitHub says the Copilot app reads the enterprise's existing managed-settings.json. The cloud agent applies applicable managed settings and only uses approved plugins and marketplaces. For supported keys, the enterprise-managed value takes precedence over a developer's local setting. Bypass-prompt controls apply only to interactive clients: the app, CLI, and VS Code.
For a first deployment, enterprise owners can create copilot/managed-settings.json in an enterprise .github-private repository, add policy keys, and commit them to the default branch. GitHub says supported clients generally apply updates within about an hour, immediately after a developer restarts a client or signs in again; the cloud agent observes changes on the next task assignment. Organizations should test the actual delay in their own environment.
The central change is not a new agent capability. It is broader governance coverage. When the same team can start an agent from an IDE, desktop app, CLI, or cloud session, any surface outside the policy can become an exception path for plugins, external marketplaces, or high-privilege commands.
Other controls still matter. Enterprises should continue checking repository permissions, pull-request review, CI checks, secret scanning, MCP tools, network egress, and audit logs. Central policy can reduce configuration drift, but it cannot prove that an agent's changes are safe or replace human review of sources and outputs.
The update shows Copilot governance moving from configuring one client to distributing policy across clients as agents spread across more work surfaces. The useful adoption tests are whether every entry point is covered, how long changes take to apply, and whether unexpected behavior can be disabled and traced quickly.
