---
title: "Turn on automatic approval"
description: "Let a separate AI safety review approve Pioneer AI actions within each technician's permissions, for the people who opt in."
url: "https://support.outpostplatform.com/pioneer-ai/automatic-approval/"
product: "depot"
type: "how-to"
status: "draft"
ai_summary: "Owners (ai:manage) set Automatic approval on Settings > AI: Off (default), Changes only, or Changes and high-impact actions. Technicians opt in from the profile menu; a second factor is needed when high-impact actions are included. A separate AI safety review may approve; it never denies, never exceeds permissions, and never approves actions on more than 25 endpoints. All logged."
availability:
  plans: ["home-lab", "essentials", "enterprise", "msp"]
  permission: "ai:manage"
  preview: false
source: "https://gitlab.com/outpostplatform/docs/-/edit/main/src/content/docs/pioneer-ai/automatic-approval.md"
license: "CC BY 4.0"
---

# Turn on automatic approval

Automatic approval lets a separate AI safety review approve the actions
Pioneer AI proposes, so a technician working through a routine fix does not
have to approve every step. It is off by default. An owner decides which kinds
of action it may approve, and each technician decides whether it applies to
them. This page shows how to turn it on, what the review checks, and what it
never does.

## Before you begin

- Choosing the mode needs `ai:manage`, which the owner holds. Opting in needs
  only `ai:use`.
- Opting in while high-impact actions are included asks for your second
  factor, so you need an authenticator app on your account. See
  [Manage your console account](/depot/account-security/).

## How it works

When Pioneer AI proposes an action that the mode covers, for a technician who
has opted in, a separate AI request reviews it before anyone is asked. The
review sees only the technician's own recent messages in the conversation and
the summary of the proposed action that the approval card would show. It
never sees Pioneer AI's replies, tool results, or
[instructions](/pioneer-ai/instructions/), so text that Pioneer AI read from an
endpoint or a ticket cannot argue for its own approval.

If your organization or the action's client has a
[safety review policy](/pioneer-ai/instructions/#how-the-safety-review-uses-a-policy),
or a team memory a person saved applies, the review checks the action a second
time against those rules, using the current time in the technician's time
zone. Both checks must allow it. These rules can only make the review
stricter.

The review answers allow or deny, with a short reason:

- **Allow.** The action runs, and its card shows **Approved by the AI safety
  review.**
- **Deny, or no clear answer.** The action is not refused. It waits for the
  technician on the usual approval card, which says **The AI safety review did
  not approve this on its own, so it needs you.** or **The AI safety review
  could not decide, so it needs you.**, with the reason.

## What it never does

- It never lets Pioneer AI do more than the technician's own permissions
  allow, and never touches the
  [actions that are never allowed](/pioneer-ai/approve-actions/#actions-that-are-never-allowed).
- It never refuses an action. Only a person can deny one.
- It never approves for a technician who has not opted in, or for a kind of
  action the mode does not cover.
- It never approves an action on more than 25 endpoints, or a high-impact
  action whose number of endpoints is not known in advance. Those always ask
  the technician.
- It approves a limited number of high-impact actions for one person in an
  hour. After that, the technician is asked.
- It never approves a task in which Pioneer AI
  [controls a remote session](/remote-access/pioneer-control/). The
  technician always decides.

## Choose the mode

1. Go to **Settings** > **AI**.
2. On the **Pioneer AI** card, in **Automatic approval**, choose one:

   | Option | What the review may approve |
   | --- | --- |
   | **Off: a technician confirms every change** | Nothing. This is the default. |
   | **Changes only: high-impact actions still ask** | **Change** actions, such as updating a ticket or acknowledging an alert. |
   | **Changes and high-impact actions** | **Change** and **High impact** actions, such as running a command or deploying software. |

3. Select **Save**.

Turning the mode off, or lowering it, takes effect at once for everyone. See
[Which actions ask](/pioneer-ai/approve-actions/#which-actions-ask) for the
tiers. Shell commands get their tier one command at a time, and a
**High impact** command in a technician's own terminal tab always asks the
technician. See [Shell commands](/pioneer-ai/approve-actions/#shell-commands).

## Opt in for yourself

Each technician who wants automatic approval turns it on for themselves.

1. In the header, open the profile menu.
2. Turn on **Let Pioneer auto-approve actions**. The switch is shown only when
   the mode is not **Off**.
3. If the mode includes high-impact actions, confirm your second factor in the
   **Confirm it is you** dialog.

To stop, turn the switch off. It takes effect at once.

If the second factor is not confirmed, the switch stays off with **Nothing
changed, because the second factor was not confirmed.**

## Verify

Ask Pioneer AI to do something the mode covers, such as adding a note to a
ticket. Its card shows **Approved by the AI safety review.**

Every action is logged with how it was approved:

- **Recent activity** on **Settings** > **AI** lists every Pioneer AI call
  with its **Confirmation**, and the review's verdict and reason are kept with
  the call.
- The [audit log](/depot/audit-log/) records the change itself, every change
  to the mode, and every time someone opts in or out.
- The review's own model use appears as **AI safety review** on **Settings** >
  **AI** > **Usage**. See [Review Pioneer AI usage](/pioneer-ai/usage/).

## Next steps

- [Approve Pioneer AI actions](/pioneer-ai/approve-actions/)
- [Pioneer AI data and privacy](/pioneer-ai/data-and-privacy/)