How risk policies work
When you've confirmed a risk is present in a video, enforcement is handled by a policy: a pre-configured sequence of steps that determines what happens next.
Policies run automatically once a risk is accepted, either by a reviewer or automatically. Understanding how they work is the key to predictable, consistent enforcement at scale.
What a policy does
A policy is a sequence of actions that runs when a risk is confirmed on a creator's video. It might message the creator, remove the video from paid distribution, end a collaboration, or simply log a warning.
Policies run without further input from your team unless you've configured a risk to require review.
Two rules to know:
Each risk is mapped to one policy. Multiple risks can share the same policy, so you don't need a separate policy for every risk type.
A policy cannot be switched off while risks are still assigned to it. Reassign those risks in your risk configuration first.
The policy types at a glance
Eight policies are available, ordered here from strictest to least strict. Detail on each one follows below.
Policy | What happens | Creator messaged |
|---|---|---|
End creator relationship | Removed from everything, creator blocklisted | Yes, final notice |
End linked collaboration | That collaboration ends, creator stays on your roster | Yes |
Request video removal | Held from GMV Max, re-checked after two days, escalates if still live | Yes |
Request metadata edit | Held from GMV Max, re-checked after one day, restored if fixed | Yes |
Warn creator and remove from GMV Max | Out of paid, stays live organically | Yes |
Send warning to creator | No enforcement, warning logged | Yes |
Remove from GMV Max only | Out of paid, quietly | No |
No action needed | Risk resolved, nothing actioned | No |
What you can and cannot change
For now, the set of policies is fixed, and so is the order of steps inside each one. This helps keep enforcement consistent across your brands and reviewers.
Three things are yours to set. A safety admin can change these on the policy detail page:
The wait length, on a policy that already includes a wait. For example, extending the two days on Request video removal if your creators typically need longer.
The message the creator receives, either the template on the policy or the message on an individual video before you save your review.
Whether the notify-TikTok step runs, on the policies that include one. It is switched off by default.
What cannot change: steps cannot be added, removed or reordered, and new policies cannot be created.
When multiple risks fire on the same video
When Unbox detects more than one risk on a video, only one policy runs: the strictest applicable policy wins. Strictness follows the order in the table above.
Example. A video is flagged for both Missing FTC Disclosure (mapped to Send warning to creator) and Competitor Disparagement (mapped to End linked collaboration). The system runs End linked collaboration only. The warning is not sent separately.
What the creator sees. The message comes from the stricter policy, not the lighter one. If that policy has a terse enforcement message, that's what the creator receives, even if the trigger was a minor issue alongside a more serious one. The message covers every risk confirmed on the video, not only the one whose policy ran.
Each policy in detail
End creator relationship
The most decisive outcome. When a risk is accepted, the system immediately removes the video from your campaigns, ends every active collaboration (Target or Open Plan) with this creator, adds them to your Blocklist, and sends them a final notice. There is no waiting period and no opportunity for the creator to remedy the issue.
Use this for risks where the content is unrecoverable and the relationship cannot continue, for example deeply offensive content or serious regulatory violations.
End linked collaboration
Similar to End creator relationship, but scoped to the specific collaboration (Target or Open Plan) that produced the video. The creator's other active collaborations with your brand are not affected.
Use this when the violation is serious enough to warrant removing the shoppable link and disassociating your brand, but doesn't disqualify the creator entirely.
Request video removal
The system removes the video from paid distribution via GMV Max immediately, messages the creator asking them to take the video down, then waits for a set period, two days by default and configurable, before re-checking whether the video has been removed.
If removed: the risk is resolved and the collaboration continues.
If still live: the system escalates automatically, ending the linked collaboration.
Use this when the video content itself is the problem and can't be remedied by an edit alone.
Request metadata edit
The system removes the video from paid distribution via GMV Max, messages the creator asking them to update the caption or title, then waits for a set period, one day by default and configurable, before re-checking the metadata.
If edited: the video is restored to paid distribution and the risk is resolved.
If unchanged: the system escalates, ending the linked collaboration.
Use this for issues that live in the caption or title rather than the video itself, for example a missing disclosure hashtag.
This policy can only be used for risks that can be resolved through a metadata edit, such as Missing Disclosure. It cannot be applied to other risk types.
Send warning to creator and remove from GMV Max
The system removes the video from GMV Max campaigns and sends the creator a message noting the issue. The risk is then resolved. The video remains live on TikTok but is excluded from paid amplification.
Use this when the content is acceptable to stay up but shouldn't be actively promoted, for example a video that makes a borderline claim you don't want amplified.
Send warning to creator
The system sends the creator a message noting the issue, then logs the risk as resolved. No enforcement action is taken. The video stays live and the collaboration continues.
Use this for minor or stylistic issues where you want to coach the creator without disrupting the campaign.
Remove from GMV Max only
The system removes the video from GMV Max campaigns and resolves the risk. No message is sent to the creator.
Use this when you want to remove a video from paid distribution without contacting the creator, for example where the content is fine for organic reach but doesn't meet your standards for amplification.
No action needed
Resolves the risk without taking any action and without messaging the creator.
Use this where every risk on a video is dismissed, or deliberately for a risk you want recorded but not enforced.
Quarantine
Quarantine is a protective layer that runs before any response policy, automatically, as soon as a risk is detected.
When any risk is flagged by Unbox AI, Quarantine immediately holds that video from live GMV Max campaigns. The hold does not require a reviewer to act first.
From there, a reviewer, or an auto-approved risk, makes a decision:
Dismiss the flag: the video is released back into GMV Max campaigns. No response policy runs.
Accept the risk: the hold is maintained, and the response policy configured for that risk runs as normal.
Where to find it. Quarantine is switched on or off in Configure Safety, on the Policies tab. There are two independent settings: one holds videos with risks detected by Unbox AI, and one holds videos Unbox cannot analyse.
Scope. A video removed from GMV Max stays out of every eligible GMV Max campaign, not only the one it was running in.
One prerequisite. All of this runs through your connected ads account. Connecting it is part of onboarding. Without it, holds and releases cannot reach your campaigns.
How policies and review mode interact
Whether a policy runs automatically or waits for a reviewer depends on how each individual risk is configured, not on the policy itself.
Auto-approve: when the AI flags a risk, it is automatically accepted and the policy runs immediately. No human step is required.
Require review: the risk is held in the review queue until a reviewer accepts or dismisses it. The policy only runs after acceptance.
The same policy can run in Auto-approve mode for one risk and Require review for another. For example, you might set Missing Disclosure to Auto-approve, while keeping Competitor Disparagement on Require review because it benefits from human judgement.
You configure this per risk in Configure Safety → Risks tab.