Skip to content
⌘K
InboxAsk a human

Setting up your risk configuration

Updated by Hugo

Risk Configuration is where you manage what Unbox detects for your brand and what happens when something is found.

Changes here take effect on all new content processed after you save. They don't retroactively affect past detections.

There are two tabs:

  • Risks – view and configure which risks you're monitoring for.

  • Policies – review the policies those risks trigger. See How risk policies work.

The Risks tab

The Risks tab lists every risk active on your account, grouped into categories. You can see the full set in the risk catalogue.

Each row shows:

  • Name and description – what the AI looks for.

  • Enabled or disabled – whether this risk is active for your brand.

  • Severity – how serious the risk is if detected.

  • Review mode – whether risks detected by Unbox AI are actioned automatically, or held for a reviewer.

  • Policy – which policy runs when this risk is confirmed.

Who can make changes

Changing your risk configuration requires a safety admin role, held either for specific brands or across the whole workspace. Admins edit the configuration using the Configure Risks button in the top right.

Everyone else can see the configuration, where the settings are visible but read-only. Working the review queue needs no admin rights at all.

To set up or change who has admin, contact support.

Enabling and disabling risks

Only enabled risks are monitored. If a risk is disabled, Unbox will not flag it on creator content, regardless of what the video contains.

Most risks can be freely toggled. Some are locked, either because they are mandated by Unbox as risks every brand must monitor, or because your workspace administrator has set them. Locked risks show a lock indicator and cannot be toggled.

Severity

Severity signals how seriously a risk should be treated if detected. It has four levels:

  • Critical – immediate brand or legal harm.

  • High – serious violations requiring enforcement.

  • Medium – issues that should be addressed but can wait for creator cooperation.

  • Low – minor issues, often stylistic.

These are defaults suggested when you first enable a risk. You can assign any policy to any severity: severity describes the nature of the risk, it doesn't restrict what you can do about it.

Changing severity does not retroactively reclassify past detections. It applies to future content only.

Review mode: Auto-approve or Require review

Each risk has a review mode setting that controls whether the configured policy runs automatically or waits for a human decision.

  • Auto-approve – when the AI flags this risk, it is confirmed automatically and the assigned policy runs immediately. Nothing appears in the review queue.

  • Require review – the flagged video waits in the review queue until someone accepts or dismisses it. The policy runs only after it is accepted.

Review mode is set per risk, not per policy. The same policy can run in Auto-approve mode for one risk and Require review for another.

Choosing between them. Auto-approve works well for risks with high detection confidence and clear-cut violations, for example missing disclosure hashtags or unlicensed music. Require review is better where context matters or false positives are more likely, such as unsubstantiated claims or controversial subjects.

When in doubt, start on Require review and move to Auto-approve once you're confident in the detection quality.

Policies

Each risk is assigned a default policy.

  • On a risk set to Require review, that policy is suggested to the reviewer when the risk is confirmed.

  • On a risk set to Auto-approve, that policy is enforced automatically.

For a full explanation of what each policy does, see How risk policies work.

Risks that need your input first

Some risks can't work until you've told Unbox what to look for, because detection depends on information specific to your brand:

  • The prohibited claims your legal or regulatory team have defined.

  • The keywords you've banned from creator content.

  • The competitor brands you want monitored.

  • The disclosure hashtags your campaigns require.

  • The third-party logos that shouldn't appear.

Each of these risks shows a configuration prompt when you open it in the risk editor, and stays disabled until you've provided the information and saved.

You can update these lists at any time. Changes apply to content processed after you save.

Saving your changes

Changes to the Risks tab are not applied until you save them. Closing the editor without saving discards your changes.

Custom risks

If you need to monitor for something not in Unbox's standard risk catalogue, you can request a custom risk using Request custom risk on the Risks tab.

Custom risks are subject to an additional charge. Your Unbox team can confirm whether your package includes access to them.

The setup process has four steps:

  1. About – give the risk a name and describe in plain English what the AI should detect.

  2. Examples – provide examples of content that should trigger the risk, and counter-examples of similar content that shouldn't.

  3. Configuration – set the severity, policy, and review mode.

  4. Notes – any additional context for the Unbox team.

Once submitted, the custom risk appears in your risk list as pending. It stays disabled until Unbox has reviewed and confirmed the detection accuracy.

This step exists to protect you. A poorly calibrated detector creates noise in your review queue, generates unnecessary creator friction, and erodes trust in your safety setup.

When Unbox marks the risk ready, you'll be notified and can enable it like any standard risk.