Docs / AI Pilot / Production Protection and Immutable Safeguards

Production Protection and Immutable Safeguards

Published · Last updated: May 2026 · 4 min read

Who this is for

Organization Owners and Admins setting up safety rules for production infrastructure, and anyone who wants to understand the hard limits built into AI Pilot.

What you will complete

Understand what Production Protection does, how to enable it, and learn about the built-in safeguards that cannot be overridden regardless of settings.

Before you begin

  • Owner or Admin role required to enable Production Protection.
  • Go to Settings → AI Agent → Agent Controls.

What is Production Protection?

Production Protection is a safety toggle that prevents AI Pilot from proposing or executing write actions on any server you have designated as a production server.

When enabled:

  • AI Pilot can still read metrics, view logs, and answer questions about production servers.
  • It cannot propose restarts, configuration changes, deployments, package installs, firewall changes, or any write operation on a production-tagged server.
  • This applies even if you explicitly ask it to perform a write action on a production server.

In plain language: Production Protection means "I trust the AI on dev and staging, but I want to keep the AI's hands off my production servers."

How to enable Production Protection

  1. Go to Settings → AI Agent → Agent Controls.
  2. Find the Production Protection toggle near the bottom of the list.
  3. Click the toggle to enable it. It saves immediately.
  4. Confirm the toggle shows Enabled.

How servers are marked as production

Production Protection applies to servers that are flagged as production in your server settings. To mark a server as production:

  1. Go to Servers and click the server you want to protect.
  2. Open the Overview tab.
  3. Find the environment tag or production flag for that server.
  4. Set it to Production.

Once marked, AI Pilot will treat that server as protected when Production Protection is enabled.

Immutable safeguards that cannot be disabled

Regardless of how you configure AI Pilot, several behaviors are hardcoded and cannot be changed by any toggle, role, or API call:

1. No silent destructive actions The AI cannot delete a server, site, database, or backup without an explicit approval card — and for all deletions, you must type the target name before the Allow button activates. This behavior is not configurable.

2. No approval card bypass AI Pilot cannot execute any write action without presenting an approval card to a human first. There is no "auto-approve" mode.

3. No secrets in chat output The AI cannot read or display environment variable values, API keys, or credentials stored in your organization. It can acknowledge that secrets exist but cannot retrieve their values.

4. Role boundaries are always enforced Viewer roles cannot approve write actions regardless of what the AI proposes. Only Members, Admins, and Owners can approve actions, and only within the scope their role permits.

5. Audit trail is always on Every AI action — proposed, approved, denied, or executed — is logged permanently. This cannot be disabled.

Step-by-step: verifying Production Protection is working

  1. Enable Production Protection (see above).
  2. Mark at least one server as production (see above).
  3. Go to AI Pilot and ask: "Please restart nginx on [your production server name]."
  4. The AI should respond that it cannot perform write actions on production servers and suggest you disable Production Protection or use a manual approach instead.
  5. Confirm no approval card appeared.

What success looks like

  • The AI responds to write requests on production servers with a refusal message explaining Production Protection is active.
  • For read requests on production servers, the AI responds normally with metrics and analysis.
  • The Production Protection toggle shows Enabled in Agent Controls.

Common errors and fixes

"Production Protection is enabled but the AI still proposed a write action on my production server" Cause: The server may not be tagged as production. Fix: Check the server's environment tag. Go to Servers → [server name] → Overview and confirm the production flag is set.

"I cannot find the environment/production flag on my server" Cause: Some imported servers may not have the tag applied automatically. Fix: Edit the server's settings and set the environment type to Production manually.

"I need to perform an emergency action on a production server with the AI" Recommended approach: Temporarily disable Production Protection under Settings → AI Agent → Agent Controls, perform the action with AI Pilot approval, then re-enable Production Protection immediately. Alternative: Perform the action manually without AI Pilot.

"The AI says it cannot act on a server but Production Protection is off" Cause: The server's per-server AI access level may be set to Read Only or No Access. Fix: Check Settings → AI Agent → Per-Server Access and set the server to Full Access.

Safety notes

  • Production Protection is a recommended default for all organizations running live customer-facing workloads.
  • Disabling Production Protection for an emergency should be followed by immediate re-enabling. Leaving it off is a security risk.
  • The immutable safeguards listed above cannot be removed by any configuration change. They are enforced at the platform level.

Related articles