Privilege Removal
Privilege Removal lists the privileged grants that no source has seen used for a long time, with the evidence behind each, so that standing privileges can be removed with confidence. Navigate to PAM Inventory > Privilege Removal.
Privilege Removal is available in Pro and Enterprise. Dismissing a recommendation needs a Manager or Administrator. See Licensing.
What is assessed
Each row is one account and privileged entitlement pair on an enabled, active account. Its status is one of:
| Status | Meaning |
|---|---|
| Recommended for removal | The last known use is at least the threshold ago (90 days by default) |
| Unknown | There is not enough usage data to say. This is never shown as unused. There are two kinds, which you can filter on separately: no usage data, where no source has reported any use of the grant or its account, and account in use, where the account was used recently but no source reports use of this particular grant |
| In use | Used within the threshold |
| Dismissed | A person decided the recommendation does not apply, and gave a reason |
The threshold is set in Administration > Settings > Privilege Removal Recommendations.
Evidence and confidence
Open a row to see its evidence: the date, where it came from, whether it is the grant's own date or only the account's, and everything that affected the confidence. A grant's own last-used date is used when a source reports one (AWS reports use per permission); otherwise the account's date is used, and a recently used account is unknown, because it says nothing about one particular grant.
Confidence says how much to trust a recommendation:
- It starts from how fully the date's source sees use: High for a source that sees all use of the credentials (such as AWS IAM's own data), Medium for sources that can miss some use (a scan's last logon, Entra or GCP sign-ins, a Windows logon seen by one Endpoint Sensor).
- It is lowered one step for each of: the account is vaulted (use through the vault may not be seen), the account name looks like a break-glass account (meant to sit unused), and the account is non-human (its use is often automated and unseen).
- It is never raised, and never below Low.
Inherited grants say so, and name the membership to remove instead. Recommendations are ordered by confidence, then the account's risk level, then how long it has been unused.
Per system and per identity
Filter by system or by identity to see only that system's or that person's grants. The totals at the top follow the same filters.
Acting on recommendations
On the Recommended for removal list, a Manager or Administrator can select grants and choose Create findings and tickets. Each selected grant becomes a finding with an owner (the system's owner) and a due date, and a ServiceNow or Jira ticket follows where ticketing is set up for the finding's severity: a High-confidence recommendation is a High finding, Medium is Medium, Low is Low. Only grants that are currently recommended are actioned; the result says how many were raised, were already actioned, or were skipped and why.
OrbisID never removes the access itself. Someone with the right to do so removes it at the source or through your identity governance process. When a later scan shows the account no longer holds the access (or the account is no longer active), the finding is resolved automatically and the recommendation is counted as removed. Actioned recommendations move to the Actioned list and the totals at the top show how many were actioned and how many have since been verified removed. A recommendation that was dismissed or actioned does not appear on the Recommended list.
Sending a recommendation to an identity governance tool for removal is not available; the finding and ticket are the route.
Dismissing
A Manager or Administrator can Dismiss a recommendation with a reason (for example a break-glass account). It stays dismissed across scans, is recorded in the audit log, and can be brought back with Restore from the Dismissed list.
Dormant privileged accounts
The Dormant Privileged Accounts KRI uses the most recent last-used date from any source, so an account used in AWS, GCP, Entra or on a Windows host with a sensor is not counted as dormant just because its last scan logon is old.